10.1007@978 3 030 33229 7 PDF

Miguel Felix Mata-Rivera
Roberto Zagal-Flores
Cristian Barría-Huidobro (Eds.)
Communications in Computer and Information Science 1053
Telematics and Computing

8th International Congress, WITCOM 2019
Merida, Mexico, November 4–8, 2019
Proceedings
Communications
in Computer and Information Science 1053
Commenced Publication in 2007
Founding and Former Series Editors:
Phoebe Chen, Alfredo Cuzzocrea, Xiaoyong Du, Orhun Kara, Ting Liu,
Krishna M. Sivalingam, Dominik Ślęzak, Takashi Washio, Xiaokang Yang,
and Junsong Yuan
Editorial Board Members

Simone Diniz Junqueira Barbosa
Pontifical Catholic University of Rio de Janeiro (PUC-Rio),
Rio de Janeiro, Brazil
Joaquim Filipe
Polytechnic Institute of Setúbal, Setúbal, Portugal
Ashish Ghosh
Indian Statistical Institute, Kolkata, India
Igor Kotenko
St. Petersburg Institute for Informatics and Automation of the Russian
Academy of Sciences, St. Petersburg, Russia
Lizhu Zhou
Tsinghua University, Beijing, China
More information about this series at http://www.springer.com/series/7899
Miguel Felix Mata-Rivera Roberto Zagal-Flores
• •
Cristian Barría-Huidobro (Eds.)
Telematics and Computing

8th International Congress, WITCOM 2019
Merida, Mexico, November 4–8, 2019
Proceedings
123
Editors
Miguel Felix Mata-Rivera Roberto Zagal-Flores
Instituto Politécnico Nacional Instituto Politécnico Nacional
Mexico City, Mexico Mexico City, Mexico
Cristian Barría-Huidobro
Universidad Mayor
Santiago, Chile
ISSN 1865-0929 ISSN 1865-0937 (electronic)

Communications in Computer and Information Science
ISBN 978-3-030-33228-0 ISBN 978-3-030-33229-7 (eBook)
https://doi.org/10.1007/978-3-030-33229-7
© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, expressed or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Nowadays, interest in telematics is increasing, especially as the field continues to be

immersed in various lines of research and converges with areas such as artificial
intelligence and data science. Without neglecting the high interest in climate change
and environmental issues, which cannot be alienated from the geospatial revolution,
high-quality jobs with a wide range of approaches were unarguably received, including
computer security and the Internet of Things. During WITCOM 2019, all these topics
were incorporated and reflected upon. This year, the research papers included machine
learning, augmented reality, and education in different cases studies.
The WITCOM conference attracts a large number of students, researchers, and
industrialists; in this edition, the GIS community had a broad participation providing an
opportunity for meeting and interaction between attendants. The current proceedings
contain the selected research papers. In total, 78 research papers were submitted and all
submissions underwent a thorough peer-review process. Three members of the
Program Committee reviewed each, and 31 were accepted (an acceptance rate of 40%).
Together, the documents presented here represent a set of high-quality contributions
to the literature on GIS, computing, and telematics that addresses a wide range of
contemporary issues. The conference program presented a broad set of session topics
that extend beyond the documents contained in these minutes. The materials for all
sessions are available on the conference website at www.witcom.upiita.ipn.mx and
www.witcom.org.mx.
More than 78 presentations were received on all the tracks and workshops of
WITCOM 2019, and it was a great effort to review them and join them into a consistent
program. We want to thank God and all those who contributed to this effort, especially
to the Laboratory of Geospatial Intelligence and Mobile Computing of UPIITA-IPN,
ANTACOM A.C., CENTROGEO, Parque Científico y Tecnológico de Yucatán, article
authors, session presenters, coordinators, members of the Program Committee, UPIITA
staff, and sponsors. Without them, the event would not have been so successful.
November 2019 Miguel Felix Mata-Rivera

Roberto Zagal-Flores
Cristian Barría-Huidobro
Organization
Organizing Committee
General Chair
Miguel Félix Mata-Rivera UPIITA-IPN, Mexico
Workshop Chair in Spatial Intelligence

Miriam Olivares Yale University, USA
Academic Chair
Roberto Zagal-Flores ESCOM-IPN, Mexico
Informatics Security Track

Cristian Barria-Huidobro Universidad Mayor, Chile
Local Manager
Jairo Zagal-Flores UNADM, Mexico
Posters Chair
Noe Sierra UPIITA-IPN, Mexico
Program Committee (Research Papers)

Christophe Claramunt Naval Academy Research Institute, France
Cristian Barria Universidad Mayor, Chile
Lorena Galeazzi Universidad Mayor, Chile
Claudio Casasolo Universidad Mayor, Chile
Alejandra Acuña Villalobos Universidad Mayor, Chile
Clara Burbano Unicomfacauca, Colombia
Gerardo Rubino Inria, France
Cesar Viho Irisa, France
Jose E. Gomez Université de Grenoble, France
Ken Arroyo Ohori Delft University of Technology, The Netherlands
Mario Aldape Perez CIDETEC-IPN, Mexico
Anzueto Rios Alvaro UPIITA-IPN, Mexico
Ludovic Moncla LIUPPA - UFR S&T de PAU (UPPA), France
Jose M. Lopez Becerra Hochschule Furtwangen University, Germany
Shoko Wakamiya Kyoto Sangyo University, Japan
Patrick Laube ZAUW, Switzerland
Sergio Ilarri University of Zaragoza, Spain
viii Organization
Sisi Zlatanova TU Delft, The Netherlands

Stephan Winter University of Melbourne, Australia
Stephen Hirtle University of Pittsburg, USA
Steve Liang University of Calgary, Canada
Tao Cheng University College London, UK
Willington Siabato Universidad Nacional, Colombia
Xiang Li East Normal China University, China
Andrea Ballatore University of London, UK
Carlos Di Bella INTA, Argentina
Haosheng Huang University of Zurich, Switzerland
Hassan Karimi University of Pittsburgh, USA
Luis Manuel Vilches CIC-IPN, Mexico
Victor Barrera Figueroa UPIITA-IPN, Mexico
Adrián Castañeda Galván UPIITA-IPN, Mexico
Thomaz Eduardo CINVESTAV-IPN, Mexico
Figueiredo Oliveira
Hiram Galeana Zapién Laboratorio de TI, Cinvestav Tamaulipas, Mexico
Laura Ivoone Garay SEPI-UPIITA, Mexico
Jiménez
Domingo Lara CINVESTAV-IPN, Mexico
Aldo Gustavo Orozco Lugo CINVESTAV-IPN, Mexico
Giovanni Guzman Lugo CIC-IPN, Mexico
Vladimir Luna CIC-IPN, Mexico
Omar Juarez Gambino ESCOM-IPN, Mexico
Itzama Lopez Yañez CIDETEC-IPN, Mexico
Miguel Ángel León Chávez BUAP-Facultad de Ciencias de la Computación,
Mexico
Alberto Luviano Juarez UPIITA-IPN, Mexico
Marco Antonio Moreno CIC-IPN, Mexico
Ibarra
Mario H. Ramírez Díaz CICATA-IPN, Mexico
Mario Eduardo Rivero Communication Networks Laboratory, CIC-IPN,
Angeles Mexico
Francisco Rodríguez CINVESTAV-IPN, Mexico
Henríquez
Patricio Ordaz Oliver Universidad Politécnica de Pachuca
Izlian Orea UPIITA-IPN, Mexico
Rolando Quintero Tellez CIC-IPN, Mexico
Grigori Sidorov CIC-IPN, Mexico
Miguel Jesus Torres Ruiz CIC-IPN, Mexico
Rosa Mercado ESIME UC, Mexico
Blanca Rico UPIITA-IPN, Mexico
Blanca Tovar UPIITA-IPN, Mexico
Chadwick Carreto ESCOM-IPN, Mexico
Ana Herrera UAQ, Mexico
Hugo Jimenez CIDESI, Mexico
Organization ix
José-Antonio León-Borges UQROO, Mexico

Alejandro Molina-Villegas CENTROGEO, Mexico
Néstor Darío Duque UNAL, Colombia
Méndez
Diego Muñoz Universidad Mayor, Chile
David Cordero Universidad Mayor, Chile
Jacobo Gonzalez-Leon UPIITA-IPN, Mexico
Saul Ortega Universidad Mayor, Chile
Robinson Osses Universidad Mayor, Chile
Hugo Lazcano ECOSUR, Mexico
Daniel Soto Universidad Mayor, Chile
Gomez-Balderas Jose GIPSA LAB, France
René Rodríguez-Zamora UAS, Mexico
Carolina Tripp Barba UAS, Mexico
Iliana Amabely UPSIN, Mexico
Leonor Espinoza UAS, Mexico
Diana López-Mesa Unicomfacauca, Colombia
Diana Castro ENCB-IPN, Mexico
Sponsors
ANTACOM A.C.
UPIITA-IPN
GIS LATAM
Collaborators
Alldatum Systems
Samani
Contents
GIS and Climate Change
Real Geographies in Virtual Space: A Practical Workflow

for Geovisualization with Immersive VR . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Florian Hruby, Rainer Ressl, Genghis De la Borbolla del Valle,
Oscar Rodríguez Paz, Verónica Aguilar Sierra,
and Jorge Humberto Muñoa Coutiño
Cross-Layer Architecture Applied to Geospatial Information

Communication in Embedded Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Luis Angel Camacho Campos, David Araujo Diaz,
and Jesus Alfredo Martinez Nuño
Methodology to Create Geospatial MODIS Dataset . . . . . . . . . . . . . . . . . . . 25

Geraldine Álvarez-Carranza and Hugo E. Lazcano-Hernández
Data Acquisition System for Solar Panel Analysis. . . . . . . . . . . . . . . . . . . . 34

Hernán De la Rosa, Margarita Mondragón,
and Bernardino Benito Salmerón-Quiroz
Pattern Recognition Through Empirical Mode Decomposition for

Temperature Time Series Between 1986 and 2019 in Mexico City
Downtown for Global Warming Assessment. . . . . . . . . . . . . . . . . . . . . . . . 45
Mauricio Gabriel Orozco-del-Castillo, Jorge J. Hernández-Gómez,
Gabriela Aurora Yañez-Casas, Mario Renán Moreno-Sabido,
Carlos Couder-Castañeda, Isaac Medina, Raúl Novelo-Cruz,
and Mauro Alberto Enciso-Aguilar
Moderate Resolution Imaging Spectroradiometer Products

Classification Using Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Javier Arellano-Verdejo
A Dynamic System Model Using Agile Method Techniques

Applied to Geomatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
René Rodríguez Zamora, Álvaro Peraza Garzón,
and Iliana Amabely Silva Hernández
Telematics and Electronics
Statistical Properties of Vehicle Residence Times for Fog

Computing Applications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
David Miguel Santiago, Mario E. Rivero-Angeles,
Laura I. Garay-Jiménez, and Izlian Y. Orea-Flores
xii Contents
Text and Image Transmission and Reception Using Light from LEDs
and a Light Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Sergio Sandoval-Reyes
12–18 GHz Microwave Frequency Band Microstrip Patch Antenna

Design for the Radio Implant Medical Devices Application . . . . . . . . . . . . . 110
Salvador Ricardo Meneses González
and Rita Trinidad Rodríguez Márquez
Analytics for Basic Products in Mexico . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Paul Millán and Félix Mata
Design and Development of Photovoltaic Power Meter . . . . . . . . . . . . . . . . 130

Jaime Vega Pérez, Blanca García, and Nayeli Vega García
Artificial Intelligence and Machine Learning
Clustering Methodology in Mixed Data Sets. . . . . . . . . . . . . . . . . . . . . . . . 145

Jacobo Gerardo González León and Miguel Félix Mata Rivera
American Sign Language Electromiographic Alphabet Sign Translator. . . . . . 162

Edgar-Armando Catalan-Salgado, Cristhian Lopez-Ramirez,
and Roberto Zagal-Flores
Artificial Intelligence as a Competitive Advantage

in the Manufacturing Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Juvenal Mendoza Valencia, Juan José Hurtado Moreno,
and Felipe de Jesús Nieto Sánchez
Software Engineering and Education
Mobile Application for the Support in the Learning of the Alphabet, Verbs
and Pronouns of the Mexican Sign Language Based on Augmented Reality . . . 183
A. Gordillo-Ramírez, O. Alonso-Cuevas, D. Ortega-Pacheco,
and U. Vélez-Saldaña
Augmented Reality with Swift in ARkit and Their Applications

to Teach Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Eduardo Eloy Loza Pacheco, Mayra Lorena Díaz Sosa,
and Miguel Jesús Torres Ruiz
Computer Tool Prototype for the Selections of Views to Materialize

in Data Cubes and Frequent Pattern. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Elizabeth Moreno Galván and Enrique Alfonso Carmona García
Contents xiii
Intelligent Learning Ecosystem in M-Learning Systems . . . . . . . . . . . . . . . . 213

Diana Carolina Burbano Gonzalez, Clara Lucia Burbano Gonzalez,
and Cristian Barria Huidobro
Application of Cyberphysical Systems Through Logistics as a Mediating

Component in Learning Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Andrés Felipe Córdoba U., Clara Lucia Burbano Gonzalez,
Julio E. Mejía M., and Jessica María Montilla M.
Internet of Things
IoT Botnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

Pamela Beltrán-García, Eleazar Aguirre-Anaya,
Ponciano Jorge Escamilla-Ambrosio, and Raúl Acosta-Bermejo
Traffic Light Control Through Intelligent Agents, Sensors and Arduino . . . . . 258
René Alberto Ojeda-Cepeda, Larissa Jeanette Peniche-Ruiz,
Iván Antonio Castro-Ynurreta, and Edwin Moises Pool-Couoh
A Clustering-Based Approach to Base Station Assignment

in IoT Cellular Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
Edgar Adrian Esquivel-Mendiola, Hiram Galeana-Zapién,
and Edwin Aldana-Bobadilla
Fuzzy Logic Controller for Automation of an Autonomous Irrigation

System Designed for Habanero Pepper (Capsicum Chinense Jacq.) . . . . . . . . 284
Jeni Molina-Puc, Mauricio Gabriel Orozco-del-Castillo,
Dakar Fernando Villafaña-Gamboa, Romeo Alam Gómez-Buenfil,
José Misael Guzmán-Tolosa, and Humberto Sarabia-Osorio
Internet of Things for Irrigation System . . . . . . . . . . . . . . . . . . . . . . . . . . . 294

J. M. Guzmán-Toloza, D. F. Villafaña-Gamboa, L. J. Peniche-Ruiz,
R. A. Gómez-Buenfil, J. K. Molina-Puc, and M. J. Rodríguez-Morayta
Architecture Proposal for Low-Cost Hydroponic IoT Based Systems . . . . . . . 305

Gutiérrez León Evelyn, Jorge Erik Montiel, Carreto Arellano Chadwick,
and Felipe Rolando Menchaca García
Informatics Security
Learning and Automatization of Searching in Digital Media, Based

on Concurrency, Weighing and Decay . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
David Cordero Vidal and Cristian Barria Huidobro
Security Hierarchy for the NDN Network: Description and Experimentation . . . 335
P. David Arjona-Villicaña and Alejandra G. Silva-Trujillo
xiv Contents
Methodology for Malware Scripting Analysis in Controlled Environments

Based on Open Source Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
Diego Muñoz, David Cordero, and Cristian Barría Huidobro
Validation of ICS Vulnerability Related to TCP/IP Protocol Implementation

in Allen-Bradley Compact Logix PLC Controller . . . . . . . . . . . . . . . . . . . . 355
Jaime Pavesi, Thamara Villegas, Alexey Perepechko, Eleazar Aguirre,
and Lorena Galeazzi
Identifying Components Belonging to Wireless Connectivity Security . . . . . . 365

Lorena Galeazzi Avalos, Cristian Barría Huidobro,
and Thamara Villegas Berbesi
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

GIS and Climate Change
Real Geographies in Virtual Space: A Practical
Workflow for Geovisualization
with Immersive VR
Florian Hruby1,2(&) , Rainer Ressl1,

Genghis De la Borbolla del Valle1, Oscar Rodríguez Paz1,
Verónica Aguilar Sierra1, and Jorge Humberto Muñoa Coutiño1
1
National Commission for the Knowledge and Use of Biodiversity
(CONABIO), Mexico City, Mexico
{fhruby,rressl,gborbolla,orodriguez,
vaguilar,jmunoa}@conabio.gob.mx
2
Department of Geography and Regional Research, University of Vienna,
Vienna, Austria
florian.hruby@univie.ac.at
Abstract. Propelled by the video game industries, virtual reality (VR) systems
are on the verge of entering the electronic mass media market. VR not just
allows generating imaginary worlds, but also visualizing real geographies and
possible past and future scenarios of real-world environments. However, VR
capabilities of current geographic information systems (GIS) are limited as yet,
thus challenging the transport of geodata into VR applications. Given the
shortcomings of GIS, the present article proposes a practical workflow for the
visualization of geographic information with immersive environments. Principal
work steps are 3D-modeling, composition, performance optimization and the
final dissemination on VR platforms. We will illustrate all elements of our
workflow using the example of an immersive VR application of a coral reef in
the Mexican Caribbean, thus providing both theoretical and practical back-
ground on the visualization of real-world data in VR space.
Keywords: Virtual reality Immersion Computer graphics Game engine

Geographic information system
1 Introduction
1.1 Immersive VR
Recent literature on geographic information science (GIScience) documents an
increasing interest in the application of immersive virtual reality (VR) technologies for
geovisualization matters [1–3]. This interest is based upon the assumption that
immersive VR stimulates understanding through the formation of spatial presence,
defined as the sense of “being there” in a virtually mediated environment [4]. In line
with current research [4, 5], immersion can be considered a technological quality of
media that describes the extent to which an illusion of reality is delivered to the user.

M. F. Mata-Rivera et al. (Eds.): WITCOM 2019, CCIS 1053, pp. 3–15, 2019.
https://doi.org/10.1007/978-3-030-33229-7_1
4 F. Hruby et al.
To create such an illusion, highly immersive VR relies on stereo head-mounted dis-

plays (HMD) and 6 degrees of freedom (DoF) tracking systems. Moreover, a high-
performance graphics processing unit (GPU) is required to render stereo images onto
the HMD at a suitable frame rate in order to sustain the experience of presence and to
avoid motion sickness.
Regarding the case of geovisualization, we can define several criteria an immersive
VR system needs to meet in order to deliver an illusion of geographic realities [2]:
(a) interactive and stereoscopic explorability; (b) 1:1 scale level, where users perceive a
VR representation of a real place as they would do being physically there [6]; (c) re-
alistic visualization of the geographic environment and all relevant agents regarding
both appearance and behavior.
1.2 From GIS to Immersive VR Environments

Considering the aforementioned definitions and criteria, immersive VR is challenging
geographic information systems (GIS) - a standard tool of GIScience - in several ways.
Firstly, direct compatibility between GIS and HMD is as yet limited. Secondly, GIS
data usually lacks resolution and information necessary to visualize geographic space
in a realistic manner (e.g. 1:1 scale level, 3D models of fauna and flora). Thirdly,
environmental characteristics such as atmospheric conditions (e.g. wind) or lighting can
be visualized just indirectly and in an abstract manner with GIS. Given these short-
comings, the present article presents a practical workflow for the visualization of GIS-
based data in immersive VR environments. This workflow is built upon four main steps
of (a) 3D-modeling, (b) ecosystem composition, (c) performance optimization and
(d) publication in VR-capable formats.
Figure 1 summarizes these steps, which will be discussed in further detail in the
following sections. In parallel, we will illustrate each element of our workflow using
the example of an immersive VR application of a coral reef in the Mexican Caribbean,
which has been developed as part of the Mexico VR (MXVR) project.
Fig. 1. Workflow for geovisualization with immersive VR - principal work steps: (a) modelling,
(b) composition, (c) performance optimization and (d) dissemination
Real Geographies in Virtual Space 5
We cannot fail to mention that parts of the workflow shown in Fig. 1 have been
already discussed by other authors (e.g. [7]). However, given the spatial extension and
complexity of the underlying coral reef, this text will complement former work by
several fundamental elements (e.g. ecosystem composition, collective movement,
performance optimization).
2 3D-Object Modelling
2.1 GIS Base Data
Using basic features of points, lines and polygons, GIS data represents real-world
phenomena in a categorized manner and from a bird’s eye perspective. To generate
meaning, GIS users have to interpret these features, e.g. in accordance with the map
legend. However, in real space we rather see 3D objects than 2D features. Hence, in
order to visualize GIS-based data within an immersive VR environment, points, lines
and polygons need to be translated into the most important and (proto)typical objects of
the real-world environment being represented.
Moreover, these objects are not distributed in a neutral and flat map space, but
rather integrated into physical world’s topography – information typically managed
within digital terrain models (DTM) by GIS. In the present case of a Caribbean coral
reef, high resolution bathymetry information, derived from WorldView-2 data, was
available [8] (Fig. 2, left).
Fig. 2. GIS base data used for the MXVR project
More arbitrary is the disaggregation of points, lines and areas into 3D models of
real-world elements. In our model case, classes of benthic habits (Fig. 2, right) were
translated into lists of species documented by the Mexican National Biodiversity
Information System [9] for each habitat within the study area shown in Fig. 2. Based
on both ecosystemic and aesthetic criteria, the most important species of these listings
were then selected for further 3D modelling (cf. Sect. 2.2).
6 F. Hruby et al.
2.2 Object Modelling
Object-Based Modelling. The probably most genuine approach of rebuilding a

realworld object in digital space is measuring the object itself. Structure from motion
(SfM), for instance, is a widely applied method of 3D modelling, where 3D structures
are being estimated from a set of partly overlapping 2D photos taken from the original
object [10].
Under appropriate conditions, SfM based modelling results not only in a detailed 3D
replica of the object’s form, but also provides high resolution textures of the object’s
surface. The Digital Life Project gives a series of examples on how SfM can be applied
to model even living animals in a realistic manner [11, 12]. An application of SfM-
based modelling within our coral reef VR visualization is given in Fig. 3.
Fig. 3. Object-based modelling of the Queen Conch (Lobatus gigas) with SfM - sample of input
scenes (left) and render of the resulting 3D model (right)
To achieve 2D images a 3D object can be reconstructed from, any commercial

camera system can be used (incl. built-in cameras of smartphones). Alternative
methods to generate input data for object-based modelling include X-ray Computerized
Tomography (X-ray CT) or Magnetic Resonance Imaging (MRI) [13].
Reference-Based Modelling. Less genuine but more common than object-based
modelling is the generation of digital geometries, materials and textures on the basis of
guide images taken from the real-world referent. This can be done, for instance, via
box-modelling techniques, where a geometric primitive (e.g. a cube) is iteratively
modified and subdivided.
A different approach is followed by sculpting, where a polygon- or voxel-based
mesh is digitally molded in analogy to clay in the physical world [14]. Regardless of
the particular technique applied, guide images (ideally taken in a systematic way from
side-, back-/front- and top-/bottom views) are used to realistically approximate the
referent with adequate detail.
Sculpting allows representing organic shapes in a very fine detail, while box-
modelling facilitates visualizations at a high geometric precision level (cf. Fig. 4).
Fig. 4. Box-modelling (left) vs. sculpting (right) – exemplified by models of loggerhead sea
turtle (Caretta caretta) and black iguana (Ctenosaura similis)
2.3 Animation
In Sect. 1.1 we argued that – in order to generate an illusion of geographic reality in
VR space – users must perceive the virtual representation of a real place at a level of
detail as they would do being physically there. This means not just realistically
modelling all objects and agents (i.e. the computed objects that represent the real-world
creature) of a geographic environment but also to simulate their motion sequences, that
is to animate the 3D models created in the previous step (Sect. 2.2).
Skeletal Animation. Skeletal animation is a commonly used approach to convert
static into dynamic 3D objects [15]. As indicated by the technique’s name, a skeleton-
like structure is defined for each object, representing its main parts (e.g. head, legs, and
arms of a sea turtle) by so-called bones. Transforming position, scale and orientation of
each bone in a systematic, frequent and hierarchical manner, motion patterns of the
real-world referent can be represented (Fig. 5).
Fig. 5. Animation of the loggerhead sea turtle (Caretta caretta): Skeleton of the 3D model (left)
and scenes from a resulting animation sequence (right)
8 F. Hruby et al.
Vertex Animation. Directly controlling the movement of vertices rather than bones,
the vertex animation is another widely applied 3D computer animation technique
relevant to immersive VR environments. Vertex positions are stored for typical stages
of a motion sequence (e.g. the left- and rightmost position of a fish’s caudal fin
movement) and then interpolated between these stages during the animation process.
Morph animation and pose animation (especially used to animate facial expressions)
are two subtypes of vertex animation [16].
3 Composition of the VR Environment
Having defined relevant GIS data sets (incl. a high-resolution DTM) and disaggregated
them into 3D models (Sect. 2), we can start blending these basic elements into a highly
immersive and realistic environment. While this blending process cannot be done with
the limited VR capabilities of current GIS, computer game engines (e.g. Unreal
Engine® or Unity 3D®) provide an alternative approach to close the gap between GIS
and VR. Game engines not just allow for a geographically referenced distribution of 3D
models, but also a realistic setting of ambience parameters. We shall discuss these main
steps at the composition stage of our workflow subsequently in further detail.
3.1 Topography
Since DTM contain 3D information (x/y position + z height) by definition, their
visualization in VR systems is a straightforward process whose results depend on
horizontal and vertical DTM resolution. Game engines can interpret topographical
information in several vector and raster formats. In the present project, a grayscaled
heightmap of the reef bathymetry (Fig. 2) was converted into a 3D mesh in Unreal
Engine (cf. Fig. 6).
Fig. 6. GeoIVE before (left) and after (right) ambience setting.
3.2 Static Elements

Flora, inanimate nature and built-up areas, for instance, can be considered as static
elements of geographic environments for conserving their position in space. Just as
topography data, we can import information on the distribution of static elements
directly (i.e. as georeferenced raster files) from GIS into game engine software. These
distribution maps can, then, be used as a reference for allocating all 3D-models
correctly.
We can exemplify the process of static element distribution relying once again on
the model case of a VR coral reef. In a first step, information on the benthic habitats
shown in Fig. 2 was (temporarily) mapped onto the bathymetry mesh. Since each
habitat/color on the reference map had been already translated into 3D models of the
most typical species (Sect. 2), we now distributed these models upon the bathymetry
mesh by assigning each model to the accordant habitats/colors of the reference map.
Figure 6 illustrates a result of this process: Areas shown in red in Fig. 2 refer to
coral colonies. Since Elkhorn corals (acropora palmata) are typical of our study area,
they had been modelled in 3D and were then assigned to all areas labelled in red on the
reference map.
3.3 Dynamic Agents

While static elements (e.g. seagrass, algae or corals within a reef) can be easily dis-
tributed in a realistic (i.e. georeferenced) manner, a different approach needs to be
adopted for the simulation of dynamic agents (e.g. fish). In the simplest case, solitary
agents have to avoid collision with other members of VR space. More complex is the
modelling of collective movements (e.g. of animal swarms). Here the individual agent
not just has to avoid collision, but also needs to polarize (i.e. to align with the average
heading of neighbors) and aggregate towards the average position of the collective.
A common way of visualizing dynamic agents in VR space is path animation,
where the 3D model is being attached to a predefined motion path (Fig. 7). Pro-
gramming of ruled based motion patterns is an alternative and more realistic, yet
computationally demanding approach, which allows to simulate not just movement but
also behavior (foraging; relations of predator vs. prey; etc.) [17].
Fig. 7. Path network used to animate fish swarms in VR.

10 F. Hruby et al.
3.4 Ambience
The Oxford English Dictionary defines ambience as “[t]he character and atmosphere of
a place.” [18]. Character and atmosphere of a geographic place depend, for instance, on
lighting conditions, ambient sound and atmospheric parameters. Including these ele-
ments into an immersive environment will strongly affect whether a given VR scene is
perceived as realistic or not (cf. Fig. 6: left vs. right). Computer game engines provide a
rich set of tools to simulate ambience by land and sea, and even underwater.
In our model case, we simulated color and visibility range of Caribbean waters by
post-process-volume and exponential height fog functions in Unreal Engine. Under-
water caustics were visualized with an animated material of the principal VR light
source, while the effect of waves, e.g. on seagrass, was simulated by a virtual wind
function. Suspended solids and air bubbles were represented by particle systems.
Moreover, non-visual ambience settings were implemented via stereo scuba sounds
[19].
4 Performance Optimization
Complex immersive geovisualizations, e.g. the VR coral reef discussed throughout this
paper, are often constructed with millions of static objects and thousands of dynamic
agents. All these elements have to be rendered in real time and at a high frame rate onto
the HMD to sustain an illusion of reality. To guarantee these frame rates not just on
highly specialized VR systems, but also on consumer-level gaming PCs or even mobile
VR devices, several improvement measures can be implemented along the whole
workflow.
At the stage of object modelling, different techniques have been tested within the
present project (cf. Sect. 2). However, SfM and sculpting result in significantly higher
polygon counts than box-modelling, so that we preferred the latter over the former
where possible. In the case of objects rich in detail (Fig. 4), the 3D mesh was sculpted
in a first step, and retopologized afterwards in order to reduce the sculpted model’s
complexity while conserving its details. Additionally, all objects were generated at
different levels of detail (LOD) both regarding their polygon mesh and textures.
At the stage of dynamic agents, skeletal animation techniques provide higher
control of motion patterns but require considerable amounts of processing power, since
three parameters (rotation, translation and scale) need to be calculated per bone, per
object and per frame. Texture based vertex animations, on the other hand, are more
complex to define, but more flexible to integrate into advanced processes of ecosystem
composition. For instance, vertex animated models can be used as hierarchically
instanced static mesh components, where just a single object needs to be processed by
the GPU while several instances of this object can be rendered. We use this method to
simulate the shoaling behavior of fish.
A final step of optimization is the relative and/or absolute segmentation of the VR
environment. In the case of relative segmentation, only objects visible from the user’s
current position in VR space will be rendered at different levels of detail. Absolute
segmentation pre-divides the environment into regions the user enters or leaves.
Figure 8 illustrates this approach, which we adopted for the MXVR project:
Suppose that Fig. 8 represents VR space in a map-like manner. Suppose further, a
given user is virtually located in the central hexagon 1 (H1). Only within this hexagon,
both static and dynamic elements are rendered, while neighboring hexagons (H1a-f)
and hexagon spots (H2 and H3) remain empty, since they are beyond the user’s range
of vision. Imagine the user now moving from H1 towards H1c: As soon as the user
leaves H1, all objects of H1 will be eliminated from the PC’s working memory, and all
objects of H1c will be loaded and rendered as soon as this hexagon is entered. The
interspace between neighboring hexagons shown in Fig. 8 corresponds with the visi-
bility range within the VR environment to avoid abrupt (dis)appearance of 3D objects.
For the areas between hexagon spots (e.g. between 1a-f and 2a-f) only static elements
are rendered. Users can traverse these areas or switch between different hexagon spots
via teleports.
Fig. 8. Absolute segmentation of geographic VR environments
5 Dissemination
5.1 User Interface

Interactive and stereoscopic explorability at a 6DoF level is a defining criteria of
immersive VR (cf. Sect. 1). To comply with this requirement, VR hardware relies on
positional tracking systems in order to register and translate the user’s position and
movement in physical space into congruent views and actions in VR space. To make
this translation between physical and virtual space as intuitive as possible, different
approaches of user interaction and interface have been discussed by human–computer
interaction (HCI) research (cf. [20]). This research focuses not just on possible tech-
niques of user input (e.g. gestures, voice), but also on effective metaphors on how to
map the user’s perception and position in VR (e.g. avatars, embodiment) [21].
Since the MXVR project is tailored for the Oculus Rift headset, we use a wireless
game controller as standard input device. Based on the flying vehicle control model
metaphor [22], we programmed the controller for surging (x-), swaying (y-) and
heaving (z-), as well as yawing, all at a constant speed. Rolling and pitching is only
enabled by movement of the head(set) to simplify navigation.
12 F. Hruby et al.
Moreover, a small information dashboard advises the users on their position, diving
depth and direction in VR space. Additional information on the species modelled is
provided via 3D icons giving access to fact sheets in audio and text format.
5.2 Release
Consumer analysis on VR systems indicates a still small but constantly growing
market, which is clearly dominated by the Oculus Rift® and HTC Vive ® platforms
[23]. Both HMD systems are broadly supported by game engine software. However,
when it comes about rendering stereo images on the fly, the actual bottleneck of
immersive VR is not so much the availability of a suitable HMD, but rather the
performance limits of current gaming GPUs. While these limits might be overcome – in
accordance with Moore’s law [24] – during the next years, streaming platforms provide
an alternative approach to deliver high resolution VR environments to a broad audi-
ence, shifting the aforementioned bottleneck from user’s hardware to digital
bandwidth.
The Oculus Rift HMD is the principal platform, the present project has been
developed for. NVidia’s GeForce GTX 1060® (or equal GPUs with at least 6 GB
memory) defines the minimum requirements to run the application fluently. For users
with lower GPU characteristics or no type of HMD at hand, also a low-immersive
desktop VR version of the MXVR project has been published. Both full- and low-
immersive applications are available online [25] under a CC BY-NC-ND 4.0 license.
Fig. 9. Non-stereoscopic (left: desktop) and stereoscopic (right: Oculus Rift®) renders from the
MXVR project (release candidate)
6 Conclusion
In this paper, we presented a practical workflow on how to convert GIS data into a
realistic immersive VR application. This workflow is just an early and only one pos-
sible way of bridging geographic and virtual reality. However, paths are made by
walking, and during this text we tried to take a few steps further to make this path.
These steps have proven feasible and effective (Fig. 9), but future improvements on our
workflow are foreseeable at all stages.
At the stage of geographic base data, next generation GIS may already provide
significantly higher VR capabilities, thus speeding up the information transport
between real and virtual space. ESRI’s CityEngine®, for instance, already offers plug-
ins to support GIS integration with game engines [26], indicating that GIScience may –
in line with our workflow – rather opt for an indirect connection between GIS and VR
via game engines than trying to make GIS directly compatible with HMDs.
At the stage of object modelling, currently common but time-consuming reference-
based techniques may be displaced by semi- and full-automated approaches like SfM.
Recently presented mobile devices with (stereo) pairs of cameras could convert SfM
based modelling into a common form of 3D data acquisition and communication.
At the stage of ecosystem composition, we mapped interactions between flora and
fauna of the MXVR project only at a very coarse level, e.g. in terms of path animations.
Implementing artificially intelligent (AI) behavior will allow us to realistically simulate
even higher order relationships, e.g. food chains or predator-prey-relationships, as well
as future or past environmental (e.g. climate change) scenarios [17].
At the stage of dissemination, increasingly powerful and affordable hardware could
further broaden the application range of VR technologies. For example, wireless HMDs
and also mixed-reality headsets (e.g. Microsoft Hololens®) may facilitate new forms of
interaction and interfaces between real and virtual space [27, 28], that is: between
human and computer.
References
1. Çöltekin, A., Oprean, D., Wallgrün, J.O., Klippel, A.: Where are we now? Re-visiting the
digital earth through human-centered virtual and augmented reality geovisualization
environments, 119–122 (2019). https://doi.org/10.1080/17538947.2018.1560986
2. Hruby, F., Ressl, R., De la Borbolla del Valle, G.: Geovisualization with immersive virtual
environments in theory and practice. Int. J. Digit. Earth 12(2), 123–136 (2019). https://doi.
org/10.1080/17538947.2018.1501106
3. Klippel, A., et al.: Transforming earth science education through immersive experiences:
delivering on a long held promise. J. Educ. Comput. Res. (2019). https://doi.org/10.1177/
0735633119854025
4. Skarbez, R., Brooks Jr., F.P., Whitton, M.C.: A survey of presence and related concepts.
ACM Comput. Surv. (CSUR) 50(6), 96 (2017). https://doi.org/10.1145/3134301
5. Cummings, J.J., Bailenson, J.N.: How immersive is enough? A meta-analysis of the effect of
immersive technology on user presence. Media Psychol. 19(2), 272–309 (2016). https://doi.
org/10.1080/15213269.2015.1015740
6. Hruby, F.: From third-person to first-person cartographies with immersive virtual
environments. Proc. Int. Cartogr. Assoc. 2, 44 (2019). https://doi.org/10.5194/ica-proc-2-
44-2019
7. Edler, D., Husar, A., Keil, J., Vetter, M., Dickmann, F.: Virtual Reality (VR) and open
source software: a workflow for constructing an interactive cartographic VR environment to
explore urban landscapes. Kartographische Nachrichten 68(1), 3–11 (2018)
8. Cerdeira-Estrada, et al.: Benthic habitat and bathymetry mapping of shallow waters in Puerto
Morelos reefs using remote sensing with a physics based data processing. In: IEEE
International Geoscience and Remote Sensing Symposium, pp. 4383–4386. IEEE, Munich
(2012). https://doi.org/10.1109/igarss.2012.6350402
14 F. Hruby et al.
9. Sarukhán, J., Jiménez, R.: Generating intelligence for decision making and sustainable use of
natural capital in Mexico. Curr. Opin. Environ. Sustain. 19, 153–159 (2016). https://doi.org/
10.1016/j.cosust.2016.02.002
10. Carrivick, J.L., Smith, M.W., Quincey, D.J.: Structure from Motion in the Geosciences.
Wiley, Chichester (2016)
11. Digital Life. http://digitallife3d.org. Accessed 11 June 2019
12. Bot, J.A., Irschick, D.J.: Using 3D photogrammetry to create open-access models of live
animals: 2D and 3D software solutions. In: Grayburn, J., Lischer-Katz, Z., Golubiewski-
Davis, K., Ikeshoji-Orlati, V. (eds.) 3D/VR in the Academic Library: Emerging Practices
and Trends, pp. 54–72. Council on Library and Information Resources, Arlington (2019)
13. Tran, M.H., Vu, H.M.Q.: A research on 3D model construction from 2D DICOM. In:
International Conference on Advanced Computing and Applications (ACOMP), pp. 158–
163. IEEE, Can Tho City (2016). https://doi.org/10.1109/acomp.2016.031
14. Knopf, G.K., Igwe, P.C.: Deformable mesh for virtual shape sculpting. Robot. Comput.-
Integr. Manuf. 21(4–5), 302–311 (2005). https://doi.org/10.1016/j.rcim.2004.11.002
15. Jensen, J.A., Burton, R.P.: Fourveo: integration of 4D animation into conventional 3D
animation workflows. Comput. Animat. Virtual Worlds 29(3–4), e1816 (2018). https://doi.
org/10.1002/cav.1816
16. Basori, A.H., Al Jahdali, H.M.A.: Emotional facial expression based on action units and
facial muscle. Int. J. Electr. Comput. Eng. 6(5), 2478–2487 (2016). https://doi.org/10.11591/
ijece.v6i5.12135
17. Magallanes Guijón, G., Hruby, F., Ressl, R., Aguilar Sierra, V., De la Borbolla del Valle, G.,
Rodríguez Paz, O.: Modelling of collective movement in immersive environments. In:
International Archives of the Photogrammetry, Remote Sensing & Spatial Information
Sciences, vol. XLII–4, pp. 397–402 (2018). https://doi.org/10.5194/isprs-archives-xlii-4-
397-2018
18. OED: The New Oxford Dictionary of English. Oxford University Press, Oxford (2001)
19. Hruby, F.: The sound of being there: audiovisual cartography with immersive virtual
environments. KN-J. Cartogr. Geogr. Inf. 69(1), 19–28 (2019). https://doi.org/10.1007/
s42489-019-00003-5
20. Wilk, M.P., Torres-Sanchez, J., Tedesco, S., O’Flynn, B.: Wearable human computer
interface for control within immersive VAMR gaming environments using data glove and
hand gestures. In: 2018 IEEE Games, Entertainment, Media Conference (GEM), pp. 1–9.
IEEE (2018). https://doi.org/10.1109/gem.2018.8516521
21. Ahn, S.J.G., Bostick, J., Ogle, E., Nowak, K.L., McGillicuddy, K.T., Bailenson, J.N.:
Experiencing nature: embodying animals in immersive virtual environments increases
inclusion of nature in self and involvement with nature. J. Comput.-Mediat. Commun. 21(6),
399–419 (2016). https://doi.org/10.1111/jcc4.12173
22. Duan, Q., Gong, J., Li, W., Shen, S., Li, R.: Improved cubemap model for 3D navigation in
geo-virtual reality. Int. J. Digit. Earth 8(11), 877–900 (2015). https://doi.org/10.1080/
17538947.2014.947339
23. Steam Hardware & Software Survey. https://store.steampowered.com/hwsurvey/Steam-
Hardware-Software-Survey-Welcome-to-Steam. Accessed 11 June 2019
24. Moore, G.E.: Progress in digital integrated electronics. Electron Devices Meet. 21, 11–13
(1975). https://doi.org/10.1109/N-SSC.2006.4804410
25. CONABIO Geovisualization. https://www.biodiversidad.gob.mx/v_ingles/region/geoviz.
html. Accessed 11 June 2019
26. Boulos, M., Lu, Z., Guerrero, P., Jennett, C., Steed, A.: From urban planning and emergency
training to Pokémon Go: applications of virtual reality GIS (VRGIS) and augmented reality
GIS (ARGIS) in personal, public and environmental health. Int. J. Health Geogr. 16, 7
(2017). https://doi.org/10.1186/s12942-017-0081-0
27. Keil, J., Edler, D., Dickmann, F.: Preparing the HoloLens for user studies: an augmented
reality interface for the spatial adjustment of holographic objects in 3D indoor environments.
KN-J. Cartogr. Geogr. Inf. 69, 1–11 (2019). https://doi.org/10.1007/s42489-019-00025-z
28. Satriadi, K.A., Ens, B., Cordeil, M., Czauderna, T., Willett, W.J., Jenny, B.: Augmented
reality map navigation with freehand gestures. In: IEEE Virtual Reality (2019, forthcoming)
Cross-Layer Architecture Applied
to Geospatial Information Communication
in Embedded Systems
Luis Angel Camacho Campos(&) , David Araujo Diaz,

and Jesus Alfredo Martinez Nuño
ESCOM-IPN, Juan de Dios Bátiz, Lindavista, 07738 Mexico City, Mexico

lcamachoc1600@alumno.ipn.mx
Abstract. This paper presents a high-level design for cross-layer architecture, it

was implemented using an embedded system, which aims an effective com-
munication for Geospatial Information through Wi-Fi technology, in order to
emulate an Automatic Dependent Surveillance Broadcasting (ADS-B) system.
This system shares geospatial information gathered from one or more Global
Navigation Satellite System (GNSS) through another communication mecha-
nism onboard. Since one module of ADS-B was proposed to be necessary for
the future for most of the controlled airspace. Therefore, the proposed system
presents the main advantage of providing a low-cost solution in low-
performance hardware. Also, it brings the capability of increment situational
awareness through a graphic interface. In this study a real-world deployment
was simulated in order to know the viability for development and node’s
location accuracy; results showed that ADS-B functionalities were attained
successfully by proposed implementation in mobile devices and it turned into a
location aid that could be useful in several scenarios.
Keywords: Communication Embedded systems Geospatial information
1 Introduction
Currently, air navigation is mostly supported by satellite systems; among these systems
are the Global Positioning System (GPS), the European Geostationary Navigation
Overlay Service (EGNOS), Galileo and others [1]. However, data gathered from these
systems are useful for final users such as pilots and drivers, above is caused by the lack
of proper communication systems that allow sharing information in order to know
nearby vehicles position.
According to [2], communication is one of the most important aspects to get
success in an operation where many vehicles participate. Therefore, different mecha-
nisms, systems, and approaches have been studied to provide an effective data link
between vehicles. In [3] for software reconfigurable is proposed in order to rearrange
logically communications devices on board when one or more of these systems fail. On
the other hand, in [4] an additional communication system based on WiMAX Tech-
nology was proposed, in order to attain a decentralized and unstructured mechanism

https://doi.org/10.1007/978-3-030-33229-7_2
Cross-Layer Architecture Applied to Geospatial Information Communication 17
called ait-to-air Communication (A2A Comm), through an Ad Hoc network configu-

ration. Also, the use of microsatellite terminals was proposed in [5], this solution brings
a centralized solution with a long-range operation but it has a high operational cost due
to the L-band congestion, which is the operating band for this kind of devices. Fol-
lowing, in [6] a theoretical concept about Wi-Fi Technology was proposed as an
alternative for the ADS-B system and according to the Federal Aviation Administration
(FAA), an ADS-B OUT module will be necessary for most controlled airspace [7] in
2020, therefore this work describes the development of a communication protocol
based on a cross-layer architecture to share geospatial information on a Mobile Ad Hoc
Network (MANET) using mobile low-performance devices.
ADS-B is a single-hop communication; i.e., the sent data is acquired by an ADS-B
IN module and the data is treated, however, ADS-B data flow is not considered as a
multi-hop communication, Oluwatosin, et al. [8] mention the challenges of this kind of
architecture, that they name device-to-device (D2D) communication. Hence, ADS-B
functionality could be extended and improved using MANET’s approach for multi-hop
communication.
Next section describes the development of proposed architecture, following by a
section that shows the results and their discussion, then a section of conclusions and
references.
2 Development of the Vehicle Location System (VLS)
ADS-B is divided mainly in two modules ADS-B OUT and ADS-B IN [9], first module
gathers information from GNSS or other Geospatial Information provider and then
transmit that information even if there are not nearby receivers; the second module
receives data and process it to be interpreted by a final user, in Fig. 1 general data flow
between vehicles is depicted, vehicles are going to be named nodes henceforth.
The Wi-Fi technology selection was based on the study presented in [10] where a
data link of 279 km was established as a result of this research, thence, in a smaller area
Wi-Fi turns into a more suitable alternative.
2.1 Cross-Layer Architecture Design

Traditionally, the Open System Interconnection (OSI) model is referred to explain how
communication is carried out through layers, for this reason, a comparison using this
model is depicted in Fig. 2, in order to explain how the proposed cross-layer archi-
tecture works.
Proposed VLS moves data from application, presentation and session layer to
transport layer where User Datagram Protocol (UDP) is used, then, due to Ad Hoc
configuration, a package is carried out from the transport layer to network layer, finally,
the standard 802.11 [11] that works in the data link and physical layers is responsible
for transmitting message. Thence, the receiver does the same action inversely.
18 L. A. Camacho Campos et al.
Fig. 1. Data Flow in an ADS-B system. A2A Comm does not need a centralized point forward
gathered data from GNSS.
Fig. 2. Cross-layer comparison with the OSI model.

2.2 Communication Protocol Design

Designed VLS implements a software approach protocol and it defines rules to transmit
and receive data but it is supported by designed protocols and standards on each layer.
i.e., the software manages lower layers to make transmission possible. In Fig. 3 ADS-B
OUT data flow diagram is depicted. A GNSS receptor is necessary and the most
common protocol used to get geospatial data from receptors such as GNSS is NMEA
[12] protocol that provides a text string with positioning data, visible satellites and
other information, predefined GGA and RMC strings from NMEA protocol, were used
to get positioning data, after, software concatenates battery state and node state to form
the message, then, a security standard cipher this message, finally an identifier is added
for transmission. UDP sockets are used in order to increase performance due to it does
not need verification or response messages, avoiding bandwidth consumption in
comparison with TCP sockets.
Fig. 3. ADS-B OUT data flow diagram.
For the ADS-B IN module, Wi-Fi adapter receives packages and the standard
802.11 treats them to the upper layer, socket listeners trigger the process to recognize
the received package and if it is not repeated security standard decipher the message. At
this point, and in the sender counterpart, the Advanced Encryption Standard (AES) [13]
is used for this purpose due to it belongs to symmetric key authentication mechanisms,
i.e., the receiver needs to have the same transmitter’s key (e.g., passphrase, shared
token or others), the above pursues bringing data privacy due to simple Ad Hoc
configuration does not provide authentication mechanisms such as WPA [14]. After
that, the actual information is obtained and finally, the software provides data as a
service that web application consumes to show on a web browser where digital car-
tography is deployed from a Geographic Information System (GIS). In Fig. 4 this
process is depicted.
Fig. 4. ADS-B IN data flow diagram.
2.3 Embedded System Implementation

The proposed system was implemented on a Raspberry Pi 3, due to it is a small and
low-cost single-board computer, also, it has a single-chip radio which operates under
802.11 b/g/n standard, Also, it has a 40 pin interface for peripherals and a full-size
HDMI port for displays [15]. By using the Operating System (OS) tools, the config-
uration for Ad Hoc mode was done.
A basic configuration for MANET should include the following parameters in order
to establish communication:
auto wlan0
iface wlan0 inet static
address 192.168.1.1
netmask 255.255.255.0
wireless-channel 1
wireless-essid MYADHOCNET
wireless-mode ad-hoc
Where address refers to static IP address and it has to be different for each node
while wireless-essid has to be the same for nodes in that network. After MANET was
setting up, the system was embedded using nodejs wrapper, giving the advantage of
deploying the system on any OS. Although this configuration works, it would be better
if this configuration occurs automatically since there could exist a large number of
nodes connected and due to its dynamism they can leave or join the MANET several
times, in [16] a prophet address allocation is proposed and it is a suitable option for
scale this system application.
Digital cartography was served by a Java-based GIS called GeoServer which
manages raster resources using Web Map Service (WMS) and serves them to the web
browser using HyperText Transfer Protocol (HTTP) [17]. There, all data is gathered,
and it is shown on different data layers. The final integration of all data layers is
depicted in Fig. 5, where a display allows an understandable data interpretation, thence,
Fig. 5. Actual photo of the deployed full system.
Fig. 6. Graphic interface with gathered geospatial data displayed.
it provides a georeferenced graphic way for providing situational awareness. Above is

due to the graphic way is the most used Human-Computer Interaction (HCI).
Consequently, graphic interface is composed by useful preprocessed data such as
latitude, longitude, altitude, vertical and horizontal speed, Universal Time Coordinated
(UTC), a number of visible satellite and zoom in and zoom out gadget. Also, according
to the pointer position a context menu shows neighbor nodes coordinates (Fig. 6).
3 Results and Discussion
After implementation, for avoiding possible collisions, accuracy test was simulated,
above due to there is a delay between position acquisition and information displaying
which derives in position difference, one aspect to consider for avoiding crashes.
Although there are many variables to take into account such as reflected signals [18], a
time-based accuracy factor is described starting in (1).
Dt ¼ tf ti ð1Þ
Where tf is the time where data is displayed and ti is the time where GNSS data is
received, both actions take place in different nodes. For (1), clock synchronization is
needed, so UTC could be used but its accuracy is 1 s, thence, Dt could take a value of
0, in small areas or high dynamic scenarios. In order to obtain a difference estimation
f
Dt, (1) is rewritten like a sum of times involved since acquisition to display, (2)
describe this sum.
e ¼ tp1 þ tp2 þ te
Dt ð2Þ
Where tp1 is processing delay from positioning data acquisition to transmission in

the source node, tp2 is processing delay from reception to display in the receiver node,
and te is the propagation delay. An estimation for propagation time et e is showed in (3),
assuming that known wave speed is c = 3 x (108) m/s, and distance average de between
neighbor nodes.
de
et e ¼ ð3Þ
c
Based on (3), now distance accuracy approximation can be calculated, this distance
could be represented as a geofencing which alert pilots or drivers of nearby nodes.
Therefore (2) is rewritten and in (4) this is shown.
e ¼ tp1 þ tp2 þ et e
Dt ð4Þ
Up to now, et e described in (3) allows obtaining a radius, i.e. it is the position

accuracy calculated from the time difference between the source and destination nodes.
Table 1. Radius estimation in different scenarios.

Vehicle de et e f
Dt s Ud
9
Car (populated city) 1 3:3 10 0.223035 2.77 0.61
Plane (Boeing 787) 500 1:6 106 0.223037 222.22 49.56
Fighter aircraft (Northrop F-5) 2.5 8:3 109 0.223035 277.77 61.95
Helicopter (Bell 412) 250 8:3 107 0.223036 62.77 13.99
With (5) is possible to get radius estimation Ud, where s is the speed of the node. Also,
taking into account data described in Table 1 some estimation is given [4, 19, 20].
e ðsÞ
Ud ð DtÞ ð5Þ
Using JavaScript tools the average of tp1 and tp1 were settled in 0.012858 and
0.210177 respectively after 10 simulations from data gathering to displaying. Distance
is showed in meters and time in seconds, so Ud is shown in meters.
Based on Table 1, radius Ud is different for each vehicle, while speed increments a
major geofencing radius is necessary in order to increase accuracy. According to
related work, this paper presents an augmented fencing solution which increases
avoidance for possible collisions; also, this work provides a low-cost solution, unlike
satellite terminals proposed in [5], on the other hand, proposed system is added on,
however, proposed solution in [3] could be used jointly thought different communi-
cation interfaces.
Despite using the same principle of Ad Hoc network, this work offers an easy and
cheap implementation alternative and proposed solution augments traditional ADS-B
single-hop communication, due to nodes relay the received package, becoming in a
multi-hop communication system, and it gives wider area coverage than the traditional
one. The accuracy is a stochastic value because many factors impact it, however, the
presented time-based accuracy bias is lower than their WiMax counterparts presented
in the results of [4].
4 Conclusions
Results obtained show a suitable alternative for new requirements in an increasing

aerospatial industry, although there are systems with similar capabilities, this work
approach allows a non invasive implementation and a low-cost alternative that can be
applied not only in rotary-wing aircrafts or planes with small dimensions, but also in
drones or Unmanned Aircraft Vehicles (UAV), cars, ships, and other dynamic element.
Safety could be the most important aspect in trends for unmanned transportation
since human mistakes are avoided but technology inherits those responsibilities,
improvements presented show aid for manned vehicles, therefore, it could be scalable,
in order to provide automatic navigation.
As part of future work, optimization is needed to accomplish international standards
that prove this system can be used in commercial flight, highly populated controlled
airspace, as well as, vehicular control according to safety protocols for autonomous
cars. Application scenarios are not limited, and above describe just a few ones; people
tracking, location aid for vehicles, and traffic assistance for smart cities, node networks,
pervasive computing and more. Also, this VLS could be provided with video streaming
capabilities, since bandwidth constraints are manageable from each node (e.g. in [21], a
Cross-Layer Technic is mentioned in order to allow video streaming).
Acknowledgment. We would like to express our gratitude to Consejo Nacional de Ciencia y

Tecnología (CONACYT), Instituto Politécnico Nacional (IPN) and, especially to Escuela
Superior de Cómputo (ESCOM), for supporting this research.
References
1. Groves, P.D.: Principles of GNSS, Inertial, and Multisensor Integrated Navigation System,
2nd edn. A. H, Boston (2013)
2. Mexican Air Force: private interview (2018)
3. Jayachandran, M., Manikandan, J.: Software reconfigurable state-of-the-art communication
suite for fighter aircraft. In: 2011 International Conference on Communication Systems and
Network Technologies, pp. 729–733 (2011)
4. Vengadesh, A., Gunasekaran, P.: Design and development of aircraft to aircraft commu-
nication by ad-hoc networks. In: 2015 International Conference on Circuit, Power and
Computing Technologies (2015)
5. Lee, L.-N., et al.: Micro satellite terminal-based high data rate communication for rotary
wing aircraft. In: 2014 IEEE Military Communications Conference, pp. 1698–1703 (2014)
6. Deepu, G.K. Sethuraman, N.R.: A study of various aircraft communication technologies. In:
International Conference on Communication and Electronics Systems (2017)
7. Federal Aviation Administration ADS Page. https://www.faa.gov/nextgen/programs/adsb/.
Accessed 15 Mar 2019
8. Amodu, O.A., Othman, M., Noordin, N.K., Idawaty, A.: A primer on design aspects, recent
advances, and challenges in cellular device-to-device communication. Ad Hoc Netw. 94,
101938 (2019)
9. Busyairah, S.A.: ADS-B system failure modes and model. J. Navig. 27, 995–1017 (2014)
10. Flickenger, R.: Very long-distance Wi-Fi networks. In: ACM SIGCOMM Workshop on
Networked Systems for Developing Regions (2008)
11. IEEE Standard for Information Technology. IEEE Standard 802.11 (2012)
12. NMEA data. https://www.gpsinformation.org/dale/nmea.htm. Accessed 02 Apr 2019
13. Crypto. https://nodejs.org/api/crypto.html. Accessed 20 Mar 2019
14. Kaur, R., Singh, T.P., Khajuria, V.: Security issues in vehicular ad-hoc network (VANET).
In: 2nd International Conference on Trends in Electronics and Informatics, pp. 884–889
(2018)
15. RPI 3 Model B Product. https://www.raspberrypi.org/products/raspberry-pi-3-model-b.
Accessed 10 Feb 2019
16. Zhou, H., Ni, L.M., Mutka, M.W.: Prophet address allocation for large scale MANETs. Ad
Hoc Netw. J. 1, 423–434 (2003)
17. GeoServer. http://geoserver.org. Accessed 01 Feb 2019
18. Yu, P., Yao, Y., Lou, M., Zhang, L.: Analysis and research of GNSS near ground signal
based on the dilution of precision. In: IEEE ICCC, pp. 1790–1794 (2016)
19. Excelsior. https://www.excelsior.com.mx/comunidad/2016/11/14/1128068. Accessed 29
Apr 2019
20. FAA. https://www.faa.gov/air_traffic/separation_standards/. Accessed 01 May 2019
21. Rathod, N., Dongre, N.: MANET routing protocol performance for video streaming. In:
2017 International Conference on Nascent Technologies in the Engineering Field (2017)
Methodology to Create Geospatial
MODIS Dataset
Geraldine Álvarez-Carranza1 and Hugo E. Lazcano-Hernández2(B)

1
El Colegio de la Frontera Sur, Estación para la Recepción de Información Satelital
ERIS-Chetumal, Av. Centenario Km 5.5, 77014 Chetumal, Quintana Roo, Mexico
2
Cátedras CONACYT-El Colegio de la Frontera Sur,
Chetumal, Quintana Roo, Mexico
hlazcanoh@ecosur.mx
Abstract. Training and testing of algorithms used in computing for

application in several studies, require datasets previously validated and
labeled. In the case of satellite remote sensing, there are several plat-
forms with large volumes of open source data. Aqua and Terra satel-
lite platforms have available the sensor MODIS (Moderate-Resolution
Imaging Spectroradiometer) which has available open access data for
earth observation. Despite the facilities offered by the MODIS data plat-
form, extracting data from a particular region for the construction of
useful dataset requires an arduous work that includes manual, semi-
automatic and automatic stages. The present study proposes a method-
ology for the construction of a geospatial dataset using MODIS sensor
data. This methodology has been successfully implemented in the con-
struction of dataset for the analysis of physical and biological variables
in the Caribbean Sea, highlighting its application in the monitoring of
Sargasso along the coastline of the state of Quintana Roo. Its application
can be extended to any of the data and products offered by the MODIS
sensor.
Keywords: MODIS dataset · Geospatial dataset · Sargasso · MODIS

processor
1 Introduction
Studies of earth observation are carried out with open-access optical satellite
data, highlighting the sensors AQUA-MODIS, TERRA-MODIS, and Landsat-
OLI [1–5]. Spectral and radiometric resolutions of MODIS sensor have been
sub-used since a few bands have been limited to the use of RGB images and
some indexes. For example, the MODIS sensor allows 38 bands with 12 bits
of radiometric resolution [6], which in principle mean 155648 possible values
for every pixel. Therefore, it is advisable to work with the data that composes
every band. Therefore, this study propose to build a dataset with the spectral
and radiometric MODIS data, to training, testing, and validate data science
algorithms. As a very first stage, to build the data set, it is necessary to define:
c Springer Nature Switzerland AG 2019
https://doi.org/10.1007/978-3-030-33229-7_3
26 G. Álvarez-Carranza and H. E. Lazcano-Hernández
study region, analysis time, physical or biological variables according to the

characteristics of the study and the spatial, temporal, radiometric and spectral
resolutions useful for the analysis. All the above will allow to know which satellite
platform and which sensor is suitable for the study. The methodology explained
below can be applied to all data compatible with the SeaDAS [7] software.
2 Materials and Methods

As an example of this methodology, we show the construction of the data set
that was used for the quantitative analysis of the presence/absence of Sargasso
along the coastline of Quintana Roo. However, the methodology is useful to any
other study that uses data that can be processed with SeaDAS.
2.1 Example Region

The example requires pixels from the coastal zone of the state of Quintana Roo,
which has an irregular perimeter, therefore the clipping tool does not work for
this case, therefore, select each pixel manually is necessary. Additionally, the
study requires only pixels within the aquatic zone. Figure 1 shows the study
area.
2.2 Dataset Construction

As an example, Aqua-MODIS sensor dataset was chosen. MODIS allows high
quality of spectral and radiometric resolution. Once the type of study and the
files to be used are designed, the creation of the dataset can be divided into
three main tasks: (1) Selection and download of files, (2) Selection and labeling
of pixels and (3) Processing. Both lists of data generated in stages one and two
feed the processing stage. Below some details of each stage are described.
2.3 Selection and Download of Files

According to dates and specific features of the study, at this stage, a list with
names of every files used for the study was created. In the case of MODIS files,
the Julian day and the universal time coordinated (UTC) was used to select
accurately the file to be used. To get the information and download the files the
lance-modis [8] and ocean-data [9] websites were used.
To automate the download of files, a software developed at home was used
(Simple web crawler for MODIS data ver 1.0)1 , which was fed with the names of
the files. Based on the study case [10], 80 PDS files of the Aqua-MODIS sensor
[11], were selected and grouped in two different classes: 42 files of the years 2015
and 2018 corresponding to presence of Sargasso and 38 files to 2014 and 2017
corresponding to the absence of Sargasso [12,13]. Only the files that correspond
to the study dates without clouds were selected, therefore, in total only 80 files
were used.
1
INDAUTOR registry number: 03-2018-092011022400-01.
Methodology to Create Geospatial MODIS Dataset 27
Fig. 1. Study Area: Represent by 115 pixels of 1 km2 (red squares) along some sections
of the coastal zone of Quintana Roo. The challenge is to extract the data from the
different MODIS bands for the pixels of interest. (Figure credit: Holger Weissenberger).
(Color figure online)
2.4 Selection and Labeling of the Pixels
In this stage, a list of latitude and longitude of every study pixels was built. The
list built in this stage allows knowing which data will be extracted from the files
in the processing stage. The visualization, selection, and ordering of the data
were done with SeaDAS. To validate the pixels position, Google Earth Pro [14]
software was used. Once the pixels have been selected and validated the file was
converted into a “.csv” format and finally downloaded. Figure 2 shows the pixel
selection and validation stage.
3.1 Data Processing
The file and pixel information generated in the previous stages feeds process-
ing stage. To automate the processing for the study, a software tool developed
at home was used (L2-JEP-MODIS-Processor)2 , which uses the SeaDAS soft-
ware as CORE. During this stage, the raw files go through several levels of
processing. The original raw data is re-projected, radiometric and atmospheric
2
INDAUTOR registry number: 03-2019-032810002000-01.
Fig. 2. Study Area: (a) 115 selected pixels using SeaDAS. (b) Approach to the southern
area close Mahahual using SeaDAS. (c) Approach to the southern area close Mahahual
using Google Earth.
corrected to finally obtain 36 differents reflectances corresponding to the wave-

lengths offered by MODIS sensor. Finally, it is possible to select the products to
be used in the study. For the present example, the surface reflectance (rhos) and
top atmosphere reflectance (rhot) products were used, which correspond to the
reflectances with atmospheric corrections. In summary all rhos and rhot values
are atmospherically corrected reflectances in two different situations: presence
and absence of Sargassum. MODIS-Aqua was the data source to calculate all
reflectances values. Figure 3 shows some examples of rhos imagery used for this
example.
To developed this study the differences between the values of the reflectances
with the presence and absence of Sargassum was compared. The latitude and
longitude values allow located the corresponding reflectance value on the earth’s
surface. The area of study is the same but the values could be different for every
file. As an example, in Fig. 4 pixels from 718 to 728 were shown, every pixel is
related to a specific area. For every file, the values related to every pixel (area),
could be different. The pixel values for different files feed the dataset.
As an example of presence and absence of Sargasso along the coast line is
depicted in Fig. 5a and b respectively. The difference in the color between the
sea with presence and with the absence of Sargasso is indeed the difference in
reflectance between both conditions and that’s just what the satellite sensor
measures.
Fig. 3. Sample of Aqua-MODIS imagery used to build the dataset. rhos product at
different wavelengths: (a) 412 nm. (b) 469 nm. (c) 555 nm. (d) 645 nm. (e) 859 nm. (f)
1240 nm. (g) 2130 nm and (h) Reference map.
Fig. 4. Pixels from 718 to 728 were shown, every pixel is related to a specific area.
For every file, the values related to every pixel could be different. The pixel values for
different files (dates) feed the dataset.
Fig. 5. Presence and absence of Sargasso along the coast line is depicted in a and b
respectively. The difference in the color between the sea with presence and with the
absence of Sargasso is indeed the difference in reflectance between both conditions and
that’s just what the sensor measures. (Photographs credit: Hugo Lazcano) (Color figure
online)
Using this methodology, the first dataset for Quintana Roo coastline with
rhos and rhot reflectance values was built. The dataset values are labeled with
presence/absence of Sargasso between 2014 and 2018 with latitude and longi-
tude data for each pixel (from 21.496124 Latitude, −87.546677 Longitude, to
18.477211 Latitude, −88.293625 Longitude), bordering the coast of Quintana
Roo. The area is located where massive arrivals of Sargassum were recorded in
2015 and 2018. This dataset allowed us to train, test and validate the ERISNet
algorithm, which is an algorithm based on deep learning techniques that allow
assigning a probability of presence/absence of Sargasso to each of the pixels used
for the study. ERISNet was successful and finally published [10].
Table 1. Dataset features.
Clases 2 Presence/Absence
Wavelengths 7 412, 469, 555, 645, 859, 1240 and 2130 nm
Attributes 14 7 rhos and 7 rhot
Instance with Sargasso 2306 Reflectance values with Sargasso
Instance no Sargasso 2209 Reflectance values without Sargasso
Total instance 4515 Total rows
Total data 63210 Total reflectance values
The Dataset3 is available for download and can be used for training and
testing other types of algorithms. The dataset is a multivariable one, real and
useful for classification. Additional features are shown in Table 1.
3
DOI: 10.7717/peerj.6842/supp-1.
Fig. 6. Dataset example, some values corresponding to rhos reflectance at different

wavelengths (412, 469, 555, 645, 859, 1240, 2130 nm). The columns correspond to the
wavelengths, the rows to the instances and the cells to the reflectance values.
The built dataset was used successfully in the training, testing, and validation
of ERISNet, an algorithm based on convolutional and recurrent neural networks
to classify pixels with presence/absence of Sargasso [10]. Figure 6 shows a dataset
example of some values corresponding to rhos.
As summary Fig. 7 shows some details of every stage of the proposal method-
ology: (1) selection and download of files (stages with light yellow fill), (2) selec-
tion and labeling of pixels (stages with light rose fill) and (3) processing (white).
The software “Simple web crawler for MODIS data ver 1.0” and L2-JEP-MODIS-
Processor were developed at home using batch and python (Anaconda 3) lan-
guages. Lenovo Workstation (with Intel Xeon EP processor, 64 GB of RAM and
Ubuntu 18.04 64 bits) was used for software developed.
Fig. 7. Process Flow Diagram: Selection and download of files (light yellow), selection
and labeling of pixels (light rose) and processing (white) (Color figure online)
4 Conclusions
This study shows the methodology to build a geospatial dataset using as input
Aqua-MODIS data. The dataset it is valuable in itself. Therefore was registered
in INDAUTOR with the name of CaribeMX-Dataset4 . The dataset is composed
of reflectance values at seven different wavelengths (412, 469, 555, 645, 859, 1,240
and 2,130 nm.) from two different MODIS products (rhos and rhot). According
to fieldwork the dataset values was labeled as Sargasso or No Sargasso (2306 rows
related to presence of Sargasso and 2209 rows related to absence of Sargasso).
Finally the dataset has 63210 values.
4
INDAUTOR registry number:03-2019-010810340300-01.
The built dataset was used successfully in the training, testing, and valida-
tion of ERISNet algorithm. The data set is available for their download and
can be applied to other studies. The proposed methodology can be used in the
construction of different dataset from other areas and for other objectives.
According to the data source and the final application, the method to build
geospatial dataset could be different. There is no better method than another,
the study emphasizes the use of open-source data and tools.
Acknowledgments. Geraldine Álvarez thanks ECOSUR for her research assistant

fellowship. We thank the NASA Ocean Biology Processing Group (OBPG) for pro-
viding all MODIS raw data used in this study. Finally the authors are grateful to the
reviewers for their contribution to improve this manuscript.
References
1. Nathan, F.: Simulating transport pathways of pelagic Sargasso from the Equatorial
Atlantic into the Caribbean Sea. Prog. Ocean. 165, 205–214 (2018). https://doi.
org/10.1016/j.pocean.2018.06.009
2. Schell, J.M., Goodwin, D.S., Siuda, A.N.: Recent Sargassum inundation events in
the Caribbean: shipboard observations reveal dominance of a previously rare form.
Oceanography 28(3), 8–11 (2015)
3. Gower, J.: Satellite images suggest a new Sargassum source region in 2011. Remote
Sens. Lett. 4(8), 764–73 (2013)
4. Wang, M.: Mapping and quantifying Sargassum distribution and coverage in the
Central West Atlantic using MODIS observations. Remote Sens. Environ. 15(183),
350–67 (2016)
5. Wang, M.: Predicting Sargassum blooms in the Caribbean Sea from MODIS obser-
vations. Geophys. Res. Lett. 44(7), 3265–73 (2017)
6. Garcı́a-Mora, T., Jean-Francois, M.: Aplicaciones del sensor MODIS para el mon-
itoreo del territorio, 1st edn. Secretarı́a de Medio Ambiente y Recursos Naturales
(Semarnat), México (2011)
7. SeaDAS. https://seadas.gsfc.nasa.gov/. Accessed June 2019
8. Lance MODIS. https://lance-modis.eosdis.nasa.gov/. Accessed June 2019
9. Ocean Color. https://oceandata.sci.gsfc.nasa.gov/MODIS-Aqua. Accessed June
2019
10. Arellano-Verdejo, J.: ERISNet: deep neural network for Sargasso detection along
the coastline of the Mexican Caribbean. PeerJ, 7, e6842, 1–19. (2019). https://doi.
org/10.7717/peerj.6842
11. MODIS-Aqua, M.-A.O.C.: NASA goddard space flight center, ocean ecology lab-
oratory, ocean biology processing group. Moderate-resolution imaging spectrora-
diometer (MODIS) Aqua l0 Data; NASA OB.DAAC, Greenbelt, MD, USA (2018).
Accessed 01 Nov 2018
12. Van Tussenbroek, B.: Severe impacts of brown tides caused by Sargasso spp. On
near-shore caribbean seagrass communities. Mar. Pollut. Bull. 122(1–2), 272–281
(2017)
13. Rodrı́guez-Martı́nez, R.E.: Afluencia masiva de sargazo pelágico a la costa del
Caribe mexicano (2014–2015). In: Florecimientos algales nocivos en México,
CICESE, Ensenada, Mexico, pp. 352–365 (2016)
14. Google earth. https://www.google.com/intl/es/earth/. Accessed June 2019
Data Acquisition System for Solar Panel
Analysis
Hernán De la Rosa1(&), Margarita Mondragón2,

and Bernardino Benito Salmerón-Quiroz1(&)
1
ESIME Azcapotzalco, Instituto Politécnico Nacional,
Avenida de las Granjas No. 682, 02250 Mexico City, Mexico
dlrgh@hotmail.com, bsalmeron@ipn.mx
2
CIIDIR Oaxaca, Instituto Politécnico Nacional, Hornos 1003,
Santa Cruz Xoxocotlán, 71230 Oaxaca, Mexico
Abstract. This project introduces a data acquisition system for solar panel
technologies, mainly for analysis and report purposes. The measured variables
are the current and voltage generated by the panel so that the power and voltage
curve can be plotted to analyze. The system is made up of two parts, the
hardware that will be designed using an Arduino circuit board with a voltage
sensor, a current sensor and a temperature sensor that feedback the modeled
circuit based on Matlab and Simscape platforms. As an addition to the hardware
system, an embedded system is designed using operational amplifiers to simu-
late a solar panel, this emulation is based on the mathematical model that gives
our values for variables generated by solar technologies. For simulation of the
modeling and plotting of the generated data, we propose the use of Wahba’s
method, for smoothing noisy data using spline function. This model will serve
as a research/experimentation tool for technologies that improve the efficiency
range of a solar cell.
Keywords: Solar energy Embedded system Matlab/Simscape DATA

network Validation parameters Autoregressive model Wahba’s problem
1 Introduction
In the last 15 years, there has been an unprecedented change in the consumption of
energy resources. The high growth in the renewable energy market, in terms of
investment, new capacities and high growth rates in developing countries, has changed
the landscape for the energy sector. We have seen the growth of unconventional
resources and improvements in the evolution of technology for all forms of energy
resources [1]. In particular, solar energy has proven to meet up this energy scenario [2].
Solar cells, also called photovoltaic (PV) cells, convert solar radiation directly into
electricity. They capture photons by exciting electrons through the band gap of a
semiconductor, which creates electron pairs and holes that are then separated by
charge, usually by PN junctions introduced by doping. The bonding interface conducts
the electrons in one direction and the holes in the other, which creates a potential
difference in the external electrodes equal to the band gap.

https://doi.org/10.1007/978-3-030-33229-7_4
Data Acquisition System for Solar Panel Analysis 35
The concept and configuration are similar to those of a semiconductor diode, except
that the electrons and holes are introduced into the junction by the excitation of the
photon and are removed at the electrodes [3]. Currently solar cells reach a wide variety
of applications and contribute to the massive or isolated generation of electrical energy.
Crystalline silicon is a well-studied material and its technology is mature, it is the most
common material used in Photo Voltaic (PV) technologies, the global production of
solar cells increases rapidly and the amount of silicon used exceeds the set of semi-
conductor devices such as diodes, transistors and circuits.
The huge gap between the potential of solar energy and its use is due to the cost and
conversion capacity. Fossil fuels satisfy our energy demands at lower cost than solar
alternatives, in part because fossil fuel deposits are concentrated sources of energy,
while the sun distributes photons quite evenly on Earth at a more modest energy
density. The use of biomass as a fuel is limited by the production capacity of the
available land and water. The limitations of cost and capacity in the use of solar energy
are addressed more effectively with a single research objective: to increase conversion
efficiency in a cost-effective manner.
Several mathematical models have been developed that describe the behavior of a
solar cell. All the models are based on the current-voltage ratio resulting from a diode,
whose manufacture represents the n-type and p-type doped semiconductors, used in the
manufacture of a solar cell [4]. These models can be used in the design and devel-
opment of data acquisition (DAQ) systems which measure physical phenomenon such
as voltage, current, temperature, pressure, or sound converting them into digital signals.
These signals in turn, are processed and conditioned by a computer with programmable
software to allow their easy analysis and visualization.
In this work a circuit is designed in Simscape to simulate the behavior of a real PV
panel. A sensor board was constructed to collect the current, voltage and power values
generated by the photovoltaic panel. The DAQ system was completed by a computer
code programed to process and smooth the signals into a format that could be used for
analysis purposes.
2 Process and Development

2.1 Mathematical Modeling of a Solar Panel
A solar cell is the building block of a solar panel. A photovoltaic module is formed by
connecting many solar cells in series and parallel. Generally, the equivalent electrical
Fig. 1. Electrical representation of a solar cell

36 H. De la Rosa et al.
circuit of a solar cell consists of a photo-current source generated by the action of solar
radiation, a diode, a resistance in parallel and a shunt resistance (representing the
internal resistance of the material to the flow of the current), as shown in Fig. 1.
Applying Kirchhoff’s law to the representation of a solar cell as seen in Fig. 1. The
next equation can be obtained.
I ¼ IPH ID ISH ð1Þ
Where IPH is the photo-generated current, ID is the current of the diode, ISH is the
shunt resistance loss in parallel and I is the output current of the cell. The first term can
be determined from next equation and depends on the irradiance, area and temperature
in which the cell is being exposed.
G
IPH ¼ ½A JSC þ aJSC ðT TSTC Þ ð2Þ
GSTC
where G is the irradiance value under given conditions, GSTC is the irradiance value
under standard conditions, A is the area of the cell, JSC is the short-circuit current
density of the cell, aJSC is the temperature coefficient of the short-circuit current
density, T is the temperature under given conditions, TSTC is the temperature under
standard conditions.
The second term from Eq. (1) is the current of the diode which can be determined
from the Shockley equation.
VD
ID ¼ IO enVT 1 ð3Þ
Where IO is the saturation current, VD is the voltage of the diode, n is equal to the
ideality factor of the diode and VT is the thermal voltage. From Eq. (3) the diode
saturation current is obtained, which is that part of the reverse current in a semicon-
ductor diode caused by diffusion of minority carriers from the neutral regions to the
depletion region
JSC A VROC
IO ¼ VOC SH ð4Þ
e 1
nVT
Where JSC is the short-circuit current density of the cell, A is the area of the cell,
VOC is the open circuit voltage of the cell, RSH is the resistance in parallel, VT is the
thermal voltage and n is equal to the ideality factor of the diode.
The third term from Eq. (1) is shunt resistance loss current which corresponds to
the current leakage is usually suppressed in some models because the parallel resistance
(RSH) usually has a very high value with respect to the series resistance.
V þ RS I
IRSH ¼ ð5Þ
RSH
Where V is the output voltage of the solar cell, RS is the series resistance, RSH is
the parallel resistance and I is the resistance current in parallel.
Equations 2, 3 and 4 are substituted in Eq. 1, so the general modeling equation for
a solar cell is obtained.
V þ RS I V þR I
S
I ¼ IPH IO e nVT 1 ð6Þ
RSH
A solar panel is formed by grouping cells in series, in order to obtain the output
current (I) of a solar panel the output voltage (V) in the Eq. 6 is divided by the number
of cells (Ns) [5, 6].
!
V þR I V
NS S
NS þ RS I
I ¼ IPH IO e nVT
1 ð7Þ
RSH
3 Data Validation
To represent the mathematical equation and to model a solar panel, the Matlab and
Simulink platform tools are used, this model is visualized as a set of subsystems
connected and configured to calculate the variables needed to be analyzed.
3.1 Cell Modeling

Using the tools from Matlab and the mathematical equations, a model is built for the
PV cell; this model represents the circuit equations of a PV cell taking into account the
effects of physical and environmental parameters such as the solar radiation and cell
temperature, as shown in Fig. 2.
Fig. 2. Simulink model designed to represent a PV module.

Inside the block that represents the PV module, the values and elements from Eq. 7
are modeled in subsystems Fig. 3.
Fig. 3. The subsystems within the designed Simulink model.
The fundamental parameters related to solar cell are short circuit current (Isc), open
circuit voltage (Voc), efficiency of solar cell and fill factor. These values are measured
using the hardware circuit built in an arduino board using voltage, current and tem-
perature sensors. Short Circuit Current is the condition when the impedance is low and
it is calculated when the voltage equals to zero.
Isc occurs at the beginning of the forward-bias sweep and is the maximum current
value in the power quadrant. For an ideal cell, this maximum current value is the total
current produced in the solar cell by photon excitation. Isc is the short-circuit current of
cell at 25 °C and an Irradiance 1000 W/m2, Iph is the light generated current directly
proportional to the solar irradiation, Io is the diode saturation current of the cell varies
with the cell temperature, Irs is the reverse saturation current of a cell at a reference
temperature and a solar irradiation, Eg is the band gap energy of the semiconductor
used in the cell and Ish stands for Parallel resistance loss current, this term of the
equation that corresponds to the current leakage is usually suppressed in some models
because the parallel resistance usually has a very high value with respect to the series
resistance [6].
Modeling the Solar Panel in Simscape. To Model the PV cell in Simscape involves
the estimation of the I-V and P-V characteristics curves to emulate the real cell under
various environmental conditions. The most popular approach is to utilize the electrical
equivalent circuit, which is primarily based on diode. The Solar Cell block in
SimElectronics Fig. 4. represents a single solar cell as a parallel current source, Iph, an
exponential diode, D and a shunt resistance, Rsh that are connected in series with a
resistance Rs.
Fig. 4. Solar cell element in Simscape.
The Solar panel circuit is modeled using elements from Simscape and Simulink.
Solar cell was connected with blocks such as current sensor and voltage sensor to
measure the current and voltage across solar cell. Blocks for varying Irradiance level
and Temperature, were also connected in the above model. Rest of the blocks were the
interface between major blocks and used for plotting I-V characteristics (Fig. 5).
Fig. 5. Simscape array for a solar panel
This model circuit works with the data we acquired from the readings of our
experimental solar cell at a lab.
Parameter Estimation Methodology. Smoothing splines are well known to provide
nice curves which smooth discrete, noisy data, in this case it is proposed a system based
on parameter estimation theory which will allow the system to filter measurements
which can be considered error. In signal processing, an autoregressive (AR) model is a
representation of a type of random process; as such, it is used to describe certain time-
varying processes, the model specifies that the output variable depends linearly on its
own previous values and on a stochastic term.
The AR model is a widely used model to understand time series data. Traditionally,
the innovation noise of the AR is modeled as Gaussian. However, many time series
applications, for example, situations arising in applications of signal processing where
the time series are non-Gaussian and heavy-tailed, either due to intrinsic data gener-
ation mechanism or existence of outliers, the AR model with more general heavy tailed
innovations is preferred.
Another issue that frequently occurs in time series is missing values, due to system
data record failure or unexpected data loss. In the recent era of data deluge, many
applications collect and process time series data for inference, learning, parameter
estimation, and decision making. The AR model is a commonly used model to analyze
time series data, where observations taken closely in time are statistically dependent on
others. In an AR time series, each sample is a linear combination of some previous
observations with a stochastic innovation.
An AR model of order p, AR(p), is defined as.
Xp
yt ¼ u 0 þ i¼1
ui yti þ et ð8Þ
Where yt is the t-th observation, u0 is a constant, ui are autoregressive coefficients,

and et is the innovation associated with the t-th observation, which is assumed to be
Gaussian distributed, as a result of the linearity of the AR model, means that the
observations are also Gaussian distributed.
The method of generalized cross validation has been shown both theoretically and
by example, to be an effective method for estimating that value of the spline smoothing
parameter which minimizes the mean square error. Excellent estimates of the derivative
are also obtained in examples involving roughly 1% and 1/10 of 1% noise, Wahba
posed the problem of finding the proper orthogonal matrix A that minimizes the cost
function [6, 7].
1 h Xn 2
i
f ð xÞ ¼ l ð MesEstimated v mes ð jÞ Þ ð9Þ
2 j¼1
This paper defines the evolution of the value of the current density in the proposed
function f ð xÞ, so that the minimum error is chosen, but it takes in account the prediction
of the state x and the coefficients of value for the state l and the measures estimated
(MesEstimated = MS) at the instant k.
The motivation for this cost function is that if the vectors are error-free and the true
matrix f ð xÞtrue, is assumed to be the same for all the measurements, then
MesEstimated is equal to vmes trueð jÞ for all j and the cost function is equal to zero,
which is clearly the minimum value, for f ð xÞ equal to f ð xÞtrue.
So for this method we define f ð xÞ as our current density Ix and adapted it to get the
next formula.
1 h Xn 2
i
Ix ¼ l j¼1
ð I est I mes ð j Þ Þ ð10Þ
2
Where IX stands for the parameter estimation in which the optimal values con-
sidered for each test are, Iest for the estimated current values, Imes is for the current date
measured and n will stand for the number of readings we get each in a minute [8].
4 Results
In this section, part of the simulation and results are represented in order to show the
performance of the proposed system, from our readings captured by our circuit, this
data is inserted in the Simscape modeling.
The readings obtained from the test are fluctuating and represent a certain percent
of noise and error; we must determine the optimal value to get the plot needed to
analyze Fig. 6.
Fig. 6. I-V and power curve plotted from variable sample reading in the laboratory
Applying the Eq. 10 in each of the sampling readings from our test the optimal
value can be determined with the minimum range of error.
As presented in the Table 1 the values obtained for ISC are calculated as the optimal
and with less percentage of error.
Table 1. Data obtained from readings

Variable Magnitude Measuring Unit
Short-circuit current, ISC 00.796 A
Open-circuit voltage, VOC 0.457 V
Irradiance used for measurements, G 1000 W=m2
Quality factor, n 1.2 N/A
Series resistance, RS 0.0056 OhmX
Fig. 7. Percentage error obtained by our optimal value compared to the highest and lowest
values of variation in the test.
This percentage error calculated using the value for short circuit current stabilizes
rapidly in the range of 1% which can be interpreted as an acceptable range; the test
results we registered have an acceptable percentage error Fig. 7. This data reading
represents the values of a single solar cell which has been modeled and plotted Fig. 8.
Fig. 8. Plotting for the I-V and power curve obtained from experimentation data for a single
solar cell.
To get the power curve we have programed a code that calculates the value using
Watt’s law, so when the plotting process in Matlab starts both curves are sent to us to
analyze.
The simscape model we designed can be modified to represent a complete solar
panel, the element we configured with our readings can be connected in series or
parallel to represent the model of Solar panel we desire Fig. 9.
Fig. 9. Solar cell series connection
Depending of the number and connection type of the solar panel, a subsystem
configuration can be created to represent the PV module needed to analyze Fig. 10.
Fig. 10. Solar panel created in a subsystem
This way, the PV panel designed can be analyzed in solitary, or add more panels to
represent a PV array, the curves obtained from the readings in this solar panel can be
saved for report or analysis Fig. 11.
Fig. 11. I-V curve and power curve for a photovoltaic panel.
5 Conclusion
This paper has analyzed the development of a method for the mathematical modeling of
PV arrays, using a simulation platform and an embedded system which simulates these
systems.
The method obtains the parameters of the I–V equation by measuring the values
generated by our solar panel, which are the open circuit voltage, short-circuit current,
and maximum output power. To accomplish our main objective which is to reduce
error and represent a clean and smooth curve for analysis and report purposes, we used
the cross validation method proposed with Wahba’s problem.
Acknowledgements. The authors gratefully thank the financial support of the Secretaria de
Investigación y Posgrado del Instituto Politécnico Nacional (SIP-IPN).
References
1. World Energy Council Home page. https://www.worldenergy.org. Accessed 2019
2. Kirmani, S., Jamil, M., Akhtar, I.: Effective low cost grid-connected solar photovoltaic system
to electrify the small scale industry/commercial building. Int. J. Renew. Energy Res. 7(2),
797–806 (2017)
3. Matsumoto, Y.: Celdas Solares De Silicio: Fundamento y Tendencia. Caos Conciencia 5, 32–
44 (2011)
4. Crabtree, G.W., Lewis, N.S.: Solar energy conversion. Phys. Today 60, 37–42 (2007)
5. Villalva, M., Gazoli, J., Filho, E.: Comprehensive approach to modeling and simulation of
photovoltaic arrays. IEEE Trans. Power Electron. 24, 1198–1208 (2009)
6. Granda-Gutiérrez, E.: Modelado y simulación de celdas y paneles solares (2013)
7. Craven, P., Wahba, G.: Smoothing noisy data with spline functions. Numer. Math. 31, 377–
403 (1978)
8. Salmerón-Quiroz, B.B., Villegas-Medina, G., Guerrero-Castellanos, J., Milhacea, E.O.,
Villalobos-Martinez, R.: A quaternion approach in the estimation of the attitude of an
industrial robot (2017)
Pattern Recognition Through Empirical
Mode Decomposition for Temperature
Time Series Between 1986 and 2019
in Mexico City Downtown for Global
Warming Assessment
Mauricio Gabriel Orozco-del-Castillo1,2 , Jorge J. Hernández-Gómez2,3(B) ,

Gabriela Aurora Yañez-Casas4 , Mario Renán Moreno-Sabido1 ,
Carlos Couder-Castañeda3 , Isaac Medina3 , Raúl Novelo-Cruz1,2 ,
and Mauro Alberto Enciso-Aguilar4
1
Departamento de Sistemas y Computación,
Tecnológico Nacional de México/I.T. Mérida, Mérida, Mexico
{mauricio.orozco,e17081430}@itmerida.edu.mx, mario@itmerida.mx
2
AAAI Student Chapter at Yucatán, México (AAAIMX),
Association for the Advancement of Artificial Intelligence, Mérida, Mexico
3
Centro de Desarrollo Aeroespacial, Instituto Politécnico Nacional,
Mexico City, Mexico
{jjhernandezgo,ccouder,imedinas}@ipn.mx
4
Escuela Superior de Ingenierı́a Mecánica y Eléctrica Unidad Zacatenco,
Sección de Estudios de Posgrado e Investigación, Instituto Politécnico Nacional,
Mexico City, Mexico
g.yanezcasas@gmail.com, mencisoa@ipn.mx
Abstract. Global warming is a real threat for the survival of life on

Earth in the following 80 years. The effects of Global Warming are par-
ticularly harmful for inhabitants of very saturated urban settlements,
which is the case of Mexico City. In this work, we analyse temperature
time series from Mexico City Downtown, taken hourly between 1986
and 2019. The gaps in the time series were interpolated through the
kriging method. Then, temporal tendencies and main frequencies were
obtained through Empirical Mode Decomposition. The first frequency
mode reveals a clear increasing tendency driven by Global Warming,
which for 2019 was of 0.72 ◦ C above a 30-year baseline period mean
between 1986 and 2016. Furthermore, the shorter periods identified in
the first intrinsic mode functions are likely driven by the solar activity
periods. It remains to find the origin of the smallest identified periods in
the time series (<0.36 years).
Keywords: Empirical Mode Decomposition · Climate Change ·

Global Warming · Time series · Artificial Intelligence · Pattern
recognition · Temperature increase · Mexico City · AI · EMD

https://doi.org/10.1007/978-3-030-33229-7_5
46 M. G. Orozco-del-Castillo et al.
1 Introduction
Nowadays, we constantly face dread events due to Climate Change (CC), like
glacier melting and intensification of drought and precipitation episodes [1,2].
Since the middle of the 20th century, scientists detected a clearly increasing
tendency in the temperature records all around the world. This information
along with paleoclimatic records during the Holocene constitute an unfailing
proof of Global Warming (GW) [3,4].
GW is defined as the increment in the global mean temperature over land
and ocean, and it has a direct effect over the whole Earth’s ecosystem [4]. In
its last report, the International Panel on Climate Change (IPCC) remarks that
for the year 2100, a rise in the mean global temperature of 2.5 ◦ C (with respect
to the mean temperature in the 1961–1980 period) is expected, driving severe
consequences for mankind as well as for nature, which threatens the survival of
life at Earth in the nearby future [4,5].
GW has unleashed a very intense research activity in CC since the second half
of the 20th century. Most of the research in the field focuses on both the study of
the anthopogenic causes and consequences of related physicochemical processes,
and on providing projections or predictions of the global mean temperature
behaviour in diverse scenarios [3].
From the point of view of natural phenomena, some of the most remarkable
investigated effects of GW are on droughts [1,6], floods [2,7], permafrost/glacier
alterations [8,9], ENSO phenomenon [10,11], hurricane activity [12,13], sea level
rise [14,15], biodiversity affectations [16,17], as well as the temperature incre-
ment itself [18,19], etc. Of special importance are the impacts of GW on human
socioeconomics. There are clear indications that GW already has importantly
affected agriculture [20,21], wine production [22,23], cattle farming [24,25], wood
assets [26,27], fishing [28,29], textile industry [30,31], transportation [32,33],
manufacturing [34,35], energy [36,37], among many others.
In 2018, an estimated 55.3% of the world’s population lived in urban settle-
ments [38], in contrast to the value of 33.616% in 1960 [39]. Moreover, by 2050
it is expected to rise to 68% (up to two thirds of mankind) [40]. Therefore, the
effects of GW on urban settlements and on their dwellers are of the greatest con-
cern to scientists and governments. In this sense, several hazards for urban envi-
ronments have been detected/forecast, for instance in interurban transportation
[41], urban-based pollution [42,43], health issues in inhabitants, [44,45], urban
design plans and policies [46,47], urban growth and land use [48,49], natural
hazards in city environments [50,51], poverty and gender aspects of population
[52,53], adaptation of current urban infrastructure [54,55], governance, policies,
resilience and adaptation [56,57], feeding challenges, [58,59], sustainability under
population growth [32,60], forced migration/mobility [61,62], as well as pro-
nounced temperature increments yielding to heat waves (heat islands) [63,64],
In this sense, it is very important to focus efforts to mitigate the effects of
GW in big cities. This is the case of Mexico City, the fifth most populated in
the world in 2018 according to United Nations [38] (and the second in America,
just below Sao Paulo, Brazil), which in the last years has faced serious events
of heat waves, precipitations, as well as forest fires and pollution peaks [65]. In
Temperature Pattern Recognition in México City and Global Warming 47
this work, we tackle the rise temperature in Mexico City Downtown by means of
computational science techniques which have proved to be successful to interpret
and forecast data arisen from both natural and social sciences, particularly those
in the field of Artificial Intelligence (AI) [66]. Artificial intelligence is a collection
of methods dealing with the simulation of human intelligence on a machine, so as
to make the machine efficient to identify and use the right piece of “Knowledge”
at a given step of solving a problem [66, section 1.2]. Despite its proved capacity
to analyse data and to solve problems, AI has been barely used in GW and
CC with respect to the huge amount of research in the area. In this sense,
relevant AI techniques to assess CC and WG have been applied, like [67], fuzzy
theory [68,69], agent-based modeling [68,70], swarm optimization [71], genetic
algoritmhs [72,73], pattern recognition [74], machine learning [75,76], cellular
automata [68,77], and artificial neural networks (ANNs) [78,79].
In this paper, currently a work in progress, we implement a pattern recog-
nition system to analyse temperature time series measured near Mexico City
Downtown. The purpose is to identify periodicities and tendencies in the data
that might suggest that temperature is being driven by GW, as well as to build
local indicators to accurately guide local strategic decision-makers to design and
implement adequate policies and actions to mitigate and counteract CC. This
work is presented as follows: In Sect. 2 we present details about the region under
study and about the analysed data, while in Sect. 3 we present the applied the-
oretical background. The obtained results are presented in Sect. 4 along with an
enriching discussion, while finally in Sects. 5 and 6 we present some final remarks,
as well as future work, respectively.
2 Region Under Study

The region under study is located in the south-central part of Mexico, in the
geographically-northern part of Mexico City, which is located in the previously
called Distrito Federal state, which lies in the Valley of Mexico/Anahuac Valley.
Orographically, it is limited at north, east, west and south by Guadalupe, Santa
Catarina, de las Cruces and Ajusco mountain ranges respectively. Pollitically, it
adjoins at south with the State of Morelos and at north, east and west with the
State of Mexico.
From the 71 weather stations operated by SIMAT (SIstema de Monitoreo
ATmosférico, an state agency operated by the Mexican government) in the
Metropolitan Zone of the Valley of Mexico (MZVM), we chose the Merced (MER)
Station, the closest one to Mexico City Downtown (1.71 km from Zocalo square).
The selection of the station was based on its centricity with respect to the urban
spot (considering the metropolitan area of the City), the time of service (length
of records), the fact that the atmosphere of the north-east of the city is more
commonly polluted [80], the consistency of its records as well as its complete-
ness. The characteristics of MER station can be observed in Table 1, while its
location can be observed in Fig. 1.
The temperature records taken by SIMAT with the version and key numbers
DT001-IAD-20190523-163843 and M-K5Y2N-6THR8-KMZJ2-FYVNE respec-
tively, consist of data taken between January 1st, 1986 and April 30th, 2019,
Table 1. Features of the selected station.
Name Keyword Long (◦ ) Lat (◦ ) Alt (m) Station ID

Merced MER −99.119594 19.42461 2245 484090170127
Fig. 1. Location of the region under study.
with a sample interval of 1 h. In this way, the time series is constituted by

292,152 records, each of which consists of the following instances: date, hour,
station, parameter (temperature in this case), value and unit (Celsius). It ought
to be remarked that the obtained temperature time series has some data gaps,
which were originated when the temperature sensor at the station is under main-
tenance or when it presents failures. Thus, data gaps are interpolated through
the kriging method [81,82], which is explained in what follows. Interpolated tem-
perature time series in the referred station and sample period can be observed
in Fig. 2.
3 Methodology
In this section, we briefly describe the methods used in this paper.
3.1 Kriging Method

Kriging method is an advanced interpolation procedure that is aimed to gen-
erate an estimated surface from a set of scattered points. The method fits a
Fig. 2. Temperature time series obtained from MER station, where the values gaps
have been interpolated through the kriging method.
mathematical function to the specified points, or to the points within an specific

radius to determine the output value for each location. This algorithm assumes
that the distance or direction between the sample points indicates an spatial
correlation which can be used to generate a surface with adequate variations.
Thus, kriging is suitable when there is a directional influence in the spatially
correlated distance within the data.
In order for kriging method to weigh measured surrounding values so to cal-
culate a prediction of a location with no measurements, a geostatistical estimator
is used [3].
Kriging method has been extensively used in both natural sciences as well
as in applied mathematics and computational sciences [83,84]. The interpolated
temperature data in MER station can be observed in Fig. 2, along with the orig-
inal temperature time series. Once the data has been interpolated, we determine
tendencies and main mode components of the time series through Empirical
Mode Decomposition (EMD).
3.2 Empirical Mode Descomposition
EMD algorithm, first presented by Huang in 1998 [85], produces smooth

envelopes defined by local extrema of a sequence and subsequent subtraction
of the mean of these envelopes from an initial sequence. It requires to identify
all the local extrema connected by cubic spline lines to produce the upper and
lower envelopes [86].
This adaptive analysis method is adequate to process non-steady and non-
linear time series. The final purpose of EMD is to divide the series in modes,
the so called Intrinsic Mode Functions (IMFs) in the time domain, which is the
main difference between EMD and other analysis methods for time series such
as Fourier transform, and wavelet decomposition. As EMD always stays in the
time domain, it has the advantage of taking less processing time to yield results.
For further information on EMD, the reader is referred to [87].

EMD has been widely used in different science fields to perform operations as
recognition [88], analysis [89], filtering [90], prediction [91], etc. From the 22
computed IMFs, the first six ones can be observed in Fig. 3.
In Fig. 3, the behaviour of the first six IMFs obtained by the application of
EMD to the original temperature time series of Fig. 2, can be observed. Fur-
thermore, in Table 2 we report the identified periods in each of the 22 computed
IMFs. These periodicities were obtained as the mean of the periods between con-
secutive peaks or valleys of the IMFs. To ease the identification of such peaks or
valleys, the temporal tendency given by the first mode was subtracted from the
rest of the mode functions.
Fig. 3. Individual view of the first six Intrinsic Mode Functions (IMFs) obtained
through the application of EMD to the temperature time series in Fig. 2.
Table 2. Identified periodicities in the 22 computed IMFs through EMD.
IMF Mean period (year) IMF Mean period (year)

D2 (t) 12.9960 D13 (t) 0.0598
D3 (t) 6.5174 D14 (t) 0.0353
D4 (t) 3.8167 D15 (t) 0.0207
D5 (t) 2.5956 D16 (t) 0.0123
D6 (t) 1.4766 D17 (t) 0.0073
D7 (t) 1.2312 D18 (t) 0.0042
D8 (t) 0.9818 D19 (t) 0.0026
D9 (t) 0.5864 D20 (t) 0.0018
D10 (t) 0.336 D21 (t) 0.0013
D11 (t) 0.1785 D22 (t) 0.0009
D12 (t) 0.0986
The first extremely notable result that ought to be remarked is the very first
IMF, D1 (t), which is observed in Fig. 3a. The lowest frequency mode, D1 (t),
reveals a clear increasing tendency in the temperature from 1986 to April, 2019.
In order to have a better acquaintance of how representative is this increment,
we take the mean of the temperature records in the time series with a 30 years
baseline, i.e. splitting the data and taking the first set of records between January
1st, 1986 and January 1st, 2016. The baseline mean temperature is 16.98 ◦ C.
Actually, the tendency shown by the first mode function (D1 (t)) reached such
value in August 29th, 1997, at 10:40:59.56, so there is a clear increasing tendency
above the baseline mean temperature since mid 1997. In this sense, we can
confirm that GW has been noticeable since such date.
Furthermore, by the end of the baseline period, the temperature increase was
0.67 ◦ C while by April 30th, 2019, it was 0.72 ◦ C above the baseline mean. In
order to try to assess the future impact of GW in the region under study, we
extrapolated the data for the first frequency function (D1 (t)) in order to try
to forecast the tendency which is present in the dataset. The extrapolation is
performed by a linear algorithm. The temperatures in the meteorological station
Table 3. Forecasted values of temperature at MER station in Mexico City downtown

through the extrapolation of the first Intrinsic Mode Function (D1 (t)) in Fig. 3a through
a linear algorithm, for three key years in Climate Change according to IPCC [92].
Key year Forecast temperature (◦ C) Temperature above baseline (◦ C)

2030 17.8328 0.8491
2050 18.0709 1.0872
2100 18.6663 1.6826
under study (MER) as well as their values above the baseline mean temperature
were forecast for three key years according to IPCC: 2030, 2050 and 2100 [92].
Such predicted values can be observed in Table 3.
As it can be observed from the data shown in Table 3, the current tendency
expects an increment of 1.09 ◦ C above the baseline temperature for 2050, which
could increment to 1.68 ◦ C for 2100. Although this simple tendency is within
the SR15 GW scenario which considers as a goal to limit the temperature to
reach 1.5 ◦ C by 2050 [93] it entails disastrous consequences to the urban spot at
Mexico City as well as to its dwellers.
As for the periodicities identified in the following EMD modes in Fig. 3, and
which are shown in Table 2, we can make the following observations.
The second IMF, D2 (t) (see Fig. 3b) reveals a coarse frequency that cor-
responds to a mean period of 12.9960 years, which is almost two years above
the typical mean duration of the solar cycle, so it is possible the solar cycle
to partially drive this IMF. Furthermore, it is interesting to note that the first
minimum in such time series occurred at August 23th, 1987 (see Fig. 3b), while
the closest minimum of solar activity was reported almost one year before, at
August 28th, 1986 [94], which reinforces this hypothesis.
Table 4. Periods T of the thirty harmonics obtained from [94]. The periods are given
in years.
Harmonic Period Harmonic Period Harmonic Period

1 10.83 11 0.98 21 0.52
2 5.42 12 0.90 22 0.49
3 3.61 13 0.83 23 0.47
4 2.71 14 0.77 24 0.45
5 2.17 15 0.72 25 0.43
6 1.81 16 0.68 26 0.42
7 1.55 17 0.64 27 0.40
8 1.35 18 0.60 28 0.39
9 1.20 19 0.57 29 0.37
10 1.08 20 0.54 30 0.36
In order to identify if the periodicities found in the subsequent Intrinsic Mode

Functions (Di (t) ∀i = 3, · · · , 22) are probably driven by solar activity, they are
compared with those reported by [94,95]. The later records, which are obtained
from different sun activity indicators as Ground Level Enhancements (GLEs),
number of sunspots, etc., are a set of short term periods within the solar activity,
and can be observed in Table 4.
In this sense, it is important to note that the last EMD frequency compo-
nents, Di (t) ∀ i > 10, have shorter periods than those reported by [94,95], so
Fig. 4. IMF 22 (D22 (t)) of the EMD applied to the original temperature time series
in Fig. 2. It can be observed that due to its small enough periodicity (1.33 days), it
reproduces qualitatively well the original time series by itself.
Table 5. Comparison of the obtained periods through the application of EMD with
those reported in the literature for the solar activity [94, 95] for the coincident IMFs,
along with the relative error between both.
IMF EMD period (year) Measured period (year) Relative error (%)
D2 (t) 12.9960 10.83 20.00
D3 (t) 6.5174 5.42 20.25
D4 (t) 3.8167 3.61 5.73
D5 (t) 2.5956 2.71 4.22
D6 (t) 1.4766 1.55 4.74
D7 (t) 1.2312 1.20 2.60
D8 (t) 0.9818 0.98 0.18
D9 (t) 0.5864 0.57 2.88
D10 (t) 0.336 0.36 6.67
no match can be made. Nevertheless, they constitute themselves a good hint for
researches to investigate the existence of such very short periods in the solar
activity.
Finally, it is important to observe that the highest frequency IMF, D22 (t),
has a really small period of 0.0009 years, which is equivalent to 1.33 days. In
this sense, its resolution is high enough to almost grasp the day-night cycles. In
Fig. 4 we can observe the IMF 22. Comparing such image with the original time
series (see Fig. 2), it is clear that the 22 IMFs obtained through EMD reproduce
qualitatively well the original time series.
5 Conclusions
In this work we presented the main frequency modes in the time series of tem-
perature records from January 1st, 1986 to April 30th, 2019, with a sample
interval of 1 h, at MER station close to Mexico City’s Downtown. The detected
periods were obtained through EMD after the series were interpolated through
kriging method. The first frequency mode revealed a clear increasing tendency
of temperature in the region under study, which surpassed the baseline period
mean since August 29th, 1997. This temperature increment, most likely due to
the direct impact of GW in the studied area, reached a value of 0.72 ◦ C above
such baseline at the end of the time series, i.e. April 30th, 2019. Furthermore,
linear extrapolation of the tendency provides increments of 0.8491 ◦ C, 1.0872 ◦ C
and 1.6826 ◦ C on the key years 2030, 2050 and 2100, according to IPCC [92,93].
On the other side, when removing the effect of GW on the time series, i.e.
D1 (t), the lower frequency components of the temperature time series are most
likely driven by the solar activity. In order to better grasp this fact, in Table 5
we present a summary of the EMD obtained periods against with those reported
in the literature. In such table, we also present the relative error between both
values. It is noticeable that the greater periods (D2 (t) and D3 (t)) have relative
errors of ≈ 20%, so it remains to study which other factors may influence these
IMFs. Nevertheless, that is out of the scope of this work.
Lower periods (D4 (t) to D10 (t)) match to some reported periods in the solar
activity with reasonable errors (≈ 5%), so they are very likely to be driven by
the sun. Finally, the lowest periods reported in Table 2 (Di (t) ∀ i > 10) are to be
investigated by researchers in order identify if they are part of the several high
frequencies present in solar activity, or if they are due to other physical factors.
6 Future Work
This work, still in progress, has three research directions to follow: (1) integrate
other data into the analysis, such as population, pollution, industry, growth,
etc., factors which could also affect the temperature increase in the metropolitan
area; (2), extend this methodology to the temperature time series arisen from
other strategically located along the MZVM so to build a more robust map of
the effects of GW in a larger region; and (3) implement AI techniques to better
extrapolate the increasing temperature tendency discovered in the first frequency
mode (see Fig. 3a) to provide more realistic forecasts.
Acknowledgements. The authors want to acknowledge the Tecnológico Nacional de

México/I.T. Mérida for the financial support provided through projects 6714.19-P and
6176.19-P, as well as partial economical support by projects 20195343 and 20190136,
as well as EDI grant, provided by SIP/IPN. Some authors also thank the AAAI and
the AAAI student chapter at Yucatán, Mexico, for the support provided to perform
this work.
References
1. Nguvava, M., Abiodun, B., Otieno, F.: Projecting drought characteristics over East
African basins at specific global warming levels. Atmos. Res. 228, 41–54 (2019).
Cited By 0
2. Yadollahie, M.: The flood in Iran: a consequence of the global warming? Int. J.
Occup. Environ. Med. 10(2), 54–56 (2019)
3. Suarez Gallareta, E., Hernández Gómez, J.J., Cetzal Balam, G., Orozco del
Castillo, M., Moreno Sabido, M., Silva Aguilera, R.A.: Sistema Hı́brido Basado
en Redes Neuronales Artificiales y Descomposición Modal Empı́rica para la Evalu-
ación de la Interrelación entre la Irradiancia Solar Total y el Calentamiento Global.
Res. Comput. Sci. 147(5), 319–332 (2018)
4. Interngubernamental Panel on Climate Change: IPCC: Working Group I Contri-
bution to the IPCC Fifth Assessment Report, Climate Change 2013: The Physical
Science Basis, IPCC, AR5. Interngubernamental Panel on Climate Change (2014)
5. Spratt, D., Dunlop, I.: Existential climate-related security risk: a scenario app-
roach, Melbourne, Australia, May 2019
6. Xu, L., Chen, N., Zhang, X.: Global drought trends under 1.5 and 2 ◦ C warming.
Int. J. Climatol. 39(4), 2375–2385 (2019). Cited By 0
7. Sen, Z.: Noah and Joseph effects: floods and droughts under global warming. Int.
J. Glob. Warm. 16(3), 347–364 (2018)
8. Ding, Y., Zhang, S., Zhao, L., Li, Z., Kang, S.: Global warming weakening the
inherent stability of glaciers and permafrost. Sci. Bull. 64(4), 245–253 (2019)
9. Kotlyakov, V.M., et al.: Glacier revival and advances in the period of global warm-
ing. Dokl. Earth Sci. 481(2), 1113–1118 (2018)
10. Ying, J., Huang, P., Lian, T., Chen, D.: Intermodel uncertainty in the change
of ENSO’s amplitude under global warming: role of the response of atmospheric
circulation to SST anomalies. J. Clim. 32(2), 369–383 (2019)
11. Jiang, N., Zhu, C.: Asymmetric changes of ENSO diversity modulated by the cold
tongue mode under recent global warming. Geophys. Res. Lett. 45(22), 12506–
12513 (2018)
12. González-Alemán, J., Pascale, S., Gutierrez-Fernandez, J., Murakami, H., Gaert-
ner, M., Vecchi, G.: Potential increase in hazard from Mediterranean hurricane
activity with global warming. Geophys. Res. Lett. 46(3), 1754–1764 (2019)
13. Duan, H., Chen, D., Lie, J.: The impact of global warming on hurricane intensity.
In: IOP Conference Series: Earth and Environmental Science, vol. 199, no. 2 (2018)
14. Khan, A.: Why would sea-level rise for global warming and polar ice-melt? Geosci.
Front. 10(2), 481–494 (2019)
15. Carvalho, K., Wang, S.: Characterizing the Indian Ocean sea level changes and
potential coastal flooding impacts under global warming. J. Hydrol. 569, 373–386
(2019)
16. Tang, L., Higa, M., Tanaka, N., Itsubo, N.: Assessment of global warming impact
on biodiversity using the extinction risk index in LCIA: a case study of Japanese
plant species. Int. J. Life Cycle Assess. 23(2), 314–323 (2018)
17. Yamaura, K., Sakaue, S., Washida, T.: An assessment of global warming and biodi-
versity: CGE EMEDA analyses. Environ. Econ. Policy Stud. 19(2), 405–426 (2017)
18. Nangombe, S.S., Zhou, T., Zhang, W., Zou, L., Li, D.: High-temperature extreme
events over Africa under 1.5 and 2 ◦ C of global warming. J. Geophys. Res.: Atmos.
124(8), 4413–4428 (2019)
19. Verheyen, J., Delnat, V., Stoks, R.: Increased daily temperature fluctuations over-
rule the ability of gradual thermal evolution to offset the increased pesticide toxicity
under global warming. Environ. Sci. Technol. 53(8), 4600–4608 (2019)
20. Huo, D., et al.: Global-warming-caused changes of temperature and oxygen alter
the proteomic profile of sea cucumber Apostichopus japonicus. J. Proteomics 193,
27–43 (2019)
21. Yousefpour, R., Nabel, J., Pongratz, J.: Simulating growth-based harvest adaptive
to future climate change. Biogeosciences 16(2), 241–254 (2019)
22. Drappier, J., Thibon, C., Rabot, A., Geny-Denis, L.: Relationship between wine
composition and temperature: impact on Bordeaux wine typicity in the context of
global warming-review. Crit. Rev. Food Sci. Nutr. 59(1), 14–30 (2019)
23. Di Lena, B., Silvestroni, O., Lanari, V., Palliotti, A.: Climate change effects on cv.
Montepulciano in some wine-growing areas of the Abruzzi region (Italy). Theor.
Appl. Climatol. 136(3–4), 1145–1155 (2019)
24. Campbell, A., Becerra, T., Middendorf, G., Tomlinson, P.: Climate change beliefs,
concerns, and attitudes of beef cattle producers in the Southern Great Plains. Clim.
Change 152(1), 35–46 (2019)
25. Duru, S., Baycan, S.: Change of daily milk yield during estrous period in Holstein
cattle raised under Mediterranean climate. Trop. Anim. Health Prod. 51, 1571–
1577 (2019)
26. Geng, A., Ning, Z., Zhang, H., Yang, H.: Quantifying the climate change mitiga-
tion potential of China’s furniture sector: wood substitution benefits on emission
reduction. Ecol. Ind. 103, 363–372 (2019)
27. Salman, A., Li, Y., Bastidas-Arteaga, E.: Impact of climate change on optimal
wood pole asset management. In: Life-Cycle Analysis and Assessment in Civil
Engineering: Towards an Integrated Vision - Proceedings of the 6th International
Symposium on Life-Cycle Civil Engineering, IALCCE 2018, pp. 1727–1734 (2019)
28. Avelino, J., et al.: Survey tool for rapid assessment of socio-economic vulnerability
of fishing communities in Vietnam to climate change. Geosciences (Switzerland)
8(12), 452 (2018)
29. Sharifuzzaman, S., Hossain, M., Chowdhury, S., Sarker, S., Chowdhury, M.,
Chowdhury, M.: Elements of fishing community resilience to climate change in
the coastal zone of Bangladesh. J. Coast. Conserv. 22(6), 1167–1176 (2018)
30. Rowe, M.: Textile industry welcomes change prompted by Paris climate deal. Twist
1(77), 14–15 (2016)
31. De L’Ecluse, S.: COP21: textile industry also acts against global warming [COP
21 le textile agit aussi contre le Réchauffement]. J. du Textile 52(2275), 2–4 (2015)
32. Yang, D., Frangopol, D.: Societal risk assessment of transportation networks under
uncertainties due to climate change and population growth. Struct. Saf. 78, 33–47
(2019)
33. Markolf, S., Hoehne, C., Fraser, A., Chester, M., Underwood, B.: Transportation
resilience to climate change and extreme weather events - beyond risk and robust-
ness. Transp. Policy 74, 174–186 (2019)
34. Meinel, U., Schüle, R.: The difficulty of climate change adaptation in manufactur-
ing firms: developing an action-theoretical perspective on the causality of adaptive
inaction. Sustainability (Switzerland) 10(2), 569 (2018)
35. Carlsson Kanyama, A., Carlsson Kanyama, K., Wester, M., Snickare, L., Söderberg,
I.L.: Climate change mitigation efforts among transportation and manufacturing
companies: the current state of efforts in Sweden according to available documen-
tation. J. Clean. Prod. 196, 588–593 (2018)
36. Bastida, L., Cohen, J., Kollmann, A., Moya, A., Reichl, J.: Exploring the role
of ICT on household behavioural energy efficiency to mitigate global warming.
Renew. Sustain. Energy Rev. 103, 455–462 (2019)
37. De Cian, E., Sue Wing, I.: Global energy consumption in a warming climate.
Environ. Resour. Econ. 72(2), 365–410 (2019)
38. United Nations: The world’s cities in 2018: data booklet. Economics and Social
Affairs series (2018)
39. World Bank: Urban population (% of total)—Data (2018). https://data.
worldbank.org/indicator/sp.urb.totl.in.zs. United Nations Population Division.
World Urbanization Prospects: 2018 Revision. Accessed 9 June 2019
40. United Nations Department of Economic and Social Affairs: 68% of the world
population projected to live in urban areas by 2050, says UN—UN DESA, June
2018. https://www.un.org/development/desa/en/news/population/2018-revision-
of-world-urbanization-prospects.html. Accessed 9 June 2019
41. Landis, J., Hsu, D., Guerra, E.: Intersecting residential and transportation CO2
emissions: metropolitan climate change programs in the age of trump. J. Plan.
Educ. Res. 39(2), 206–226 (2019)
42. Trájer, A., Nagy, G., Domokos, E.: Exploration of the heterogeneous effect of cli-
mate change on ozone concentration in an urban environment. Int. J. Environ.
Health Res. 29(3), 276–289 (2019)
43. Baldermann, C., Lorenz, S.: UV radiation in Germany: influences of ozone deple-
tion and climate change and measures to protect the population. Bundesgesund-
heitsblatt - Gesundheitsforschung - Gesundheitsschutz 62(5), 639–645 (2019)
44. Orimoloye, I., Mazinyo, S., Kalumba, A., Ekundayo, O., Nel, W.: Implications of
climate variability and change on urban and human health: a review. Cities 91,
213–223 (2019)
45. Woodward, A., Baumgartner, J., Ebi, K., Gao, J., Kinney, P., Liu, Q.: Population
health impacts of China’s climate change policies. Environ. Res. 175, 178–185
(2019)
46. Aminipouri, M., et al.: Urban tree planting to maintain outdoor thermal comfort
under climate change: the case of Vancouver’s local climate zones. Build. Environ.
158, 226–236 (2019)
47. Sun, Y., Xie, S., Zhao, S.: Valuing urban green spaces in mitigating climate change:
a city-wide estimate of aboveground carbon stored in urban green spaces of China’s
capital. Glob. Change Biol. 25(5), 1717–1732 (2019)
48. Doan, V., Kusaka, H., Nguyen, T.: Roles of past, present, and future land use
and anthropogenic heat release changes on urban heat island effects in Hanoi,
Vietnam: numerical experiments with a regional climate model. Sustain. Cities
Soc. 47, 101479 (2019)
49. Chapman, S., Thatcher, M., Salazar, A., Watson, J., McAlpine, C.: The impact of
climate change and urban growth on urban climate and heat stress in a subtropical
city. Int. J. Climatol. 39(6), 3013–3030 (2019)
50. Zhang, C., Li, S., Luo, F., Huang, Z.: The global warming hiatus has faded away:
an analysis of 2014–2016 global surface air temperatures. Int. J. Climatol. 2019,
1–16 (2019)
51. Shastri, H., Ghosh, S., Paul, S., Shafizadeh-Moghadam, H., Helbich, M., Karmakar,
S.: Future urban rainfall projections considering the impacts of climate change and
urbanization with statistical-dynamical integrated approach. Clim. Dyn. 52(9–10),
6033–6051 (2019)
52. Dodman, D., Archer, D., Satterthwaite, D.: Editorial: responding to climate change
in contexts of urban poverty and informality. Environ. Urban. 31(1), 3–12 (2019)
53. Gran Castro, J., Ramos De Robles, S.: Climate change and flood risk: vulnerability
assessment in an urban poor community in Mexico. Environ. Urban. 31(1), 75–92
(2019)
54. Xiong, L., Yan, L., Du, T., Yan, P., Li, L., Xu, W.: Impacts of climate change on
urban extreme rainfall and drainage infrastructure performance: a case study in
Wuhan City, China. Irrig. Drain. 68(2), 152–164 (2019)
55. De la Sota, C., Ruffato-Ferreira, V., Ruiz-Garcı́a, L., Alvarez, S.: Urban green
infrastructure as a strategy of climate change mitigation. A case study in Northern
Spain. Urban For. Urban Green. 40, 145–151 (2019)
56. Bellinson, R., Chu, E.: Learning pathways and the governance of innovations in
urban climate change resilience and adaptation. J. Environ. Plan. Policy Manag.
21(1), 76–89 (2019)
57. Martinez-Juarez, P., Chiabai, A., Suárez, C., Quiroga, S.: Insights on urban and
periurban adaptation strategies based on stakeholders’ perceptions on hard and
soft responses to climate change. Sustainability (Switzerland) 11(3), 647 (2019)
58. Lal, R.: Managing urban soils for food security and adaptation to climate change.
In: Vasenev, V., Dovletyarova, E., Cheng, Z., Prokof’eva, T.V., Morel, J.L.,
Ananyeva, N.D. (eds.) SUITMA 2017. SG, pp. 302–319. Springer, Cham (2019).
https://doi.org/10.1007/978-3-319-89602-1 35
59. Raimundo, I.: Food insecurity in the context of climate change in Maputo city,
Mozambique: challenges and coping strategies. ROUTLEDGE in Association with
GSE Research (2016)
60. Mahaut, V., Andrieu, H.: Relative influence of urban-development strategies and
water management on mixed (separated and combined) sewer overflows in the
context of climate change and population growth: a case study in Nantes. Sustain.
Cities Soc. 44, 171–182 (2019)
61. Januszkiewicz, K., Świtoń, M.: Climate change and population mobility - envision-
ing infrastructure to reduce disaster’s impact on cities. In: International Multidis-
ciplinary Scientific GeoConference Surveying Geology and Mining Ecology Man-
agement, SGEM, vol. 17, pp. 519–526 (2017)
62. Khavarian-Garmsir, A., Pourahmad, A., Hataminejad, H., Farhoodi, R.: Climate
change and environmental degradation and the drivers of migration in the context
of shrinking cities: a case study of Khuzestan province, Iran. Sustain. Cities Soc.
47, 101480 (2019)
63. Trihamdani, A.R., Lee, H.S., Kubota, T., Iizuka, S., Phuong, T.T.T.: Urban cli-
mate challenges in Hanoi: urban heat Islands and global warming. In: Kubota,
T., Rijal, H.B., Takaguchi, H. (eds.) Sustainable Houses and Living in the Hot-
Humid Climates of Asia, pp. 529–539. Springer, Singapore (2018). https://doi.org/
10.1007/978-981-10-8465-2 48
64. Li, G., Zhang, X., Mirzaei, P., Zhang, J., Zhao, Z.: Urban heat island effect of a
typical valley city in China: responds to the global warming and rapid urbanization.
Sustain. Cities Soc. 38, 736–745 (2018)
65. Trejo-González, A., et al.: Quantifying health impacts and economic costs of PM2.5
exposure in Mexican cities of the national urban system. Int. J. Public Health
64(4), 561–572 (2019)
66. Konar, A.: Artificial Intelligence and Soft Computing: Behavioral and Cognitive
Modeling of the Human Brain. CRC Press, Boca Raton (1999)
67. Woo, T.: Artificial intelligence-based modelling for global warming analysis incor-
porated with the nuclear energy productions. Int. J. Glob. Warm. 17(4), 389–400
(2019)
68. Lu, Q., Joyce, J., Imen, S., Chang, N.B.: Linking socioeconomic development, sea
level rise, and climate change impacts on urban growth in New York City with a
fuzzy cellular automata-based Markov chain model. Environ. Plan. B: Urban Anal.
City Sci. 46(3), 551–572 (2019)
69. Woo, T.: Global warming analysis for greenhouse gases impacts comparable to
carbon-free nuclear energy using neuro-fuzzy algorithm. Int. J. Glob. Warm. 17(2),
219–233 (2019)
70. Gubareva, M., Gomes, O.: On the edge of climate change: in a search of an adequate
agent-based methodology to model environmental dynamics. In: Sequeira, T., Reis,
L. (eds.) Climate Change and Global Development. CE, pp. 37–57. Springer, Cham
(2019). https://doi.org/10.1007/978-3-030-02662-2 3
71. Atuahene, S., Bao, Y., Yevenyo Ziggah, Y., Semwaah Gyan, P., Li, F.: Short-
term electric power forecasting using dual-stage hierarchical wavelet-particle swarm
optimization-adaptive neuro-fuzzy inference system PSO-ANFIS approach based
on climate change. Energies 11(10), 2822 (2018)
72. Mohammadi, B.: “Prediction of effective climate change indicators using statistical
downscaling approach and impact assessment on pearl millet (Pennisetum glaucum
L.) yield through genetic algorithm in Punjab, Pakistan” by Asmat Ullah, Nasrin
Salehnia, Sohrab Kolsoumi, Ashfaq Ahmad, Tasneem Khaliq. Ecol. Indic. 101,
973–974 (2019)
73. Yoon, E., et al.: Modeling spatial climate change landuse adaptation with multi-
objective genetic algorithms to improve resilience for rice yield and species richness
and to mitigate disaster risk. Environ. Res. Lett. 14(2), 024001 (2019)
74. Vaghefi, S.A., Abbaspour, N., Kamali, B., Abbaspour, K.: A toolkit for climate
change analysis and pattern recognition for extreme weather conditions - case
study: California-Baja California Peninsula. Environ. Model Softw. 96, 181–198
(2017)
75. Wu, C., Chen, Y., Peng, C., Li, Z., Hong, X.: Modeling and estimating aboveground
biomass of Dacrydium pierrei in China using machine learning with climate change.
J. Environ. Manag. 234, 167–179 (2019)
76. Crane-Droesch, A.: Machine learning methods for crop yield prediction and climate
change impact assessment in agriculture. Environ. Res. Lett. 13(11), 114003 (2018)
77. Collados-Lara, A.J., Pardo-Igúzquiza, E., Pulido-Velazquez, D.: A distributed cel-
lular automata model to simulate potential future impacts of climate change on
snow cover area. Adv. Water Resour. 124, 106–119 (2019)
78. Nile, B., Hassan, W., Alshama, G.: Analysis of the effect of climate change on
rainfall intensity and expected flooding by using ANN and SWMM programs.
ARPN J. Eng. Appl. Sci. 14(5), 974–984 (2019)
79. Aslan, Z., Erdemir, G., Feoli, E., Giorgi, F., Okcu, D.: Effects of climate change
on soil erosion risk assessed by clustering and artificial neural network. Pure Appl.
Geophys. 176(2), 937–949 (2019)
80. Quezada, R., Martinez Rossier, L., Hernandez Garduno, E., Catalan Vazquez, M.,
Quezada, E., Perez Neria, J.: Comparison of carbon monoxide levels in air expired
from pedestrian in an urban area and a rural area. Rev. del Inst. Nac. de Enferm.
Respiratorias 10(1), 13–17 (1997)
81. Matheron, G.: Principles of geostatistics. Econ. Geol. 58(8), 1246–1266 (1963)
82. Krige, D.: A statistical approach to some basic mine valuation problems on the
Witwatersrand. J. Chem. Metall. Min. Soc. S. Afr. 52, 201–215 (1952)
83. Zhang, J., Li, X., Yang, R., Liu, Q., Zhao, L., Dou, B.: An extended Kriging
method to interpolate near-surface soil moisture data measured by wireless sensor
networks. Sensors 17(6), 1390 (2017)
84. Liu, L., Cheng, Y., Wang, X.: Genetic algorithm optimized Taylor Kriging surro-
gate model for system reliability analysis of soil slopes. Landslides 14(2), 535–546
(2017)
85. Huang, N.E., et al.: The empirical mode decomposition and the Hilbert spectrum
for nonlinear and non-stationary time series analysis. Proc. R. Soc. Lond. A: Math.
Phys. Eng. Sci. 454, 903–995 (1998)
86. Kim, D., Oh, H.S.: EMD: a package for empirical mode decomposition and Hilbert
spectrum. R J. 1(1), 40–46 (2009)
87. Huang, N.E.: Review of empirical mode decomposition. In: Wavelet Applications
VIII, vol. 4391, pp. 71–81. International Society for Optics and Photonics (2001)
88. Du, H.K., Cao, J.X., Xue, Y.J., Wang, X.J.: Seismic facies analysis based on self-
organizing map and empirical mode decomposition. J. Appl. Geophys. 112, 52–61
(2015)
89. Nunes, J.C., Bouaoune, Y., Delechelle, E., Niang, O., Bunel, P.: Image analysis by
bidimensional empirical mode decomposition. Image Vis. Comput. 21(12), 1019–
1026 (2003)
90. Andrade, A.O., Nasuto, S., Kyberd, P., Sweeney-Reed, C.M., Van Kanijn, F.: EMG
signal filtering based on empirical mode decomposition. Biomed. Signal Process.
Control 1(1), 44–55 (2006)
91. Drakakis, K.: Empirical mode decomposition of financial data. Int. Math. Forum
4, 1191–1202 (2008)
92. Intergubernamental Panel on Climate Change: Homepage (2014). http://www.
ipcc.ch/home languages main spanish.shtml. Accessed 01 Apr 2019
93. Masson-Delmotte, V., et al.: Global Warming of 1.5C. An IPCC Special Report
on the impacts of global warming of 1.5C above pre-industrial levels and related
global greenhouse gas emission pathways, in the context of strengthening the global
response to the threat of climate change, sustainable development, and efforts to
eradicate poverty. Technical report 1, IPCC, Geneva, Switzerland (2018, in Press)
94. Orozco-Del-Castillo, M., Ortiz-Alemán, J., Couder-Castañeda, C., Hernández-
Gómez, J., Solı́s-Santomé, A.: High solar activity predictions through an artificial
neural network. Int. J. Mod. Phys. C 28(06), 1750075 (2017)
95. Kane, R.: Short-term periodicities in solar indices. Sol. Phys. 227(1), 155–175
(2005)
Moderate Resolution Imaging
Spectroradiometer Products
Classification Using Deep Learning
Javier Arellano-Verdejo(B)
El Colegio de la Frontera Sur, Estación para la Recepción de Información Satelital

ERIS-Chetumal, Av. Centenario Km 5.5, 77014 Chetumal, Quintana Roo, Mexico
javier.arellano@mail.ecosur.mx
Abstract. During the last years, the algorithms based on Artificial

Intelligence have increased their popularity thanks to their application
in multiple areas of knowledge. Nowadays with the increase of storage
capacities and computing power, as well as the incorporation of new
technologies for massively parallel processing (GPUs and TPUs) and
Cloud Computing, it is increasingly common to incorporate this kind of
algorithms and technology in tasks with a deep social and technological
impact. In the present work a new Convolutional Neural Network spe-
cialized in the automatic classification of Moderate Resolution Imaging
Spectroradiometer satellite products is proposed. The proposed architec-
ture has shown a high-generalization by classifying more than 250,000
images with 99.99% accuracy. The methodology designed also can be
extended, with other types of images, to make detection of Sargassum,
oil spills, red tide, etc.
Keywords: Deep learning · MODIS · Remote sensing · Data science ·

Machine learning
1 Introduction
The Artificial Intelligence (AI) concept, was proposed by John McCarthy in

Dartmouth (United States) during the summer of 1956. The AI is the result
of the synergy between multiple disciplines such as computer science, mathe-
matics, and logic (to mention just a few), with the aim of getting a machine to
imitate the human cognitive functions such as learning and solving problems in
an independent way. As a result of AI, new areas of knowledge have emerged such
as: case-based reasoning, expert systems, Bayesian networks, robotics, pattern
recognition, neural networks and machine learning to mention just a few [1].
Remote sensing (RS) is based on the acquisition of information about an
object or phenomenon without having physical contact with it. The RS is used
by multiple areas of study including geography, hydrology, meteorology, oceanog-
raphy, geology, humanitarian applications, etc. The main source of information
https://doi.org/10.1007/978-3-030-33229-7_6
62 J. Arellano-Verdejo
used by the satellital RS is provided by various sensors on board space missions

such as Landsat, RADAR, Sentinel, and MODIS [2]. The Moderate-Resolution
Imaging Spectroradiometer (i.e MODIS), is an instrument of Earth observation
placed in orbit by NASA in 1999 aboard the space platform EOS AM-1 TERRA
and in 2002 on board EOS PM-1 AQUA. MODIS has a spectral resolution of
36 bands between 0.4 to 14.4 µm, has a radiometric resolution of 12 bits and a
spatial resolution of 250, 500 and 1000 m. The main use of the different bands of
the MODIS sensor is related to the detection of clouds and aerosols (bands 1 to
7), ocean color, phytoplankton and biogeochemistry (bands 8 to 16) and water
vapor (bands 16 to 19) however, the main application of the MODIS sensor is
related to the generation of the “MODIS Products”, which are used by scientists
from all over the world to carry out studies related mainly to the global change
of the planet. MODIS products can be classified into five major groups: level
1 products, atmospheric products, land products, cryosphere products, and sea
products. Recently, the combination of MODIS satellite data with data science
techniques and AI has given rise to many practical applications for Mexico such
as the detection of Sargasso on the coast of Quintana Roo [3] and the study
of the state of health of Banco Chinchorro [4]. Access to MODIS data is open
and free through NASA websites, so developing tools to manipulate, classify
and exploit them is essential for the development of algorithms and techniques
related to the handling of this type of information and its application to remote
sensing problems.
In the present work, a new AI algorithm based on a Convolutional Neural
Network is proposed for the classification of MODIS satellite products. The
proposed neural network has shown high performance and accuracy, classifying
more than 250,000 images with 99.99% of accuracy.
2 MODIS Products Classification

2.1 Artificial Neural Networks
An Artificial Neural Network (ANN) is a computational model inspired by the
interconnection and exchange of information between biological neurons of living
beings. As can be seen in Fig. 1a, an ANN is formed by a set of artificial neurons
organized in layers, in the classical model, these layers are organized into: input
layer, output layer and one or more hidden layers. The artificial neurons (see
Fig. 1b) are the minimum unit of computation of an ANN, its main function is
to compute the weighted sum between the inputs Xi and the weights Wi . The
result is processed by an activation function f which generates the final output
S. Depending on the type of network architecture, the S output can be used as
input for another neuron or simply be part of the network’s output layer [9].
Within the field of Artificial Intelligence, RNAs are part of the Machine
Learning (ML) discipline and belong to the set of supervised learning algorithms.
Supervised learning is a technique that allows generating a mathematical model
capable of correctly classifying information in different groups. It uses a set of
pairs of tagged data and, through an iterative training process, approximates the
MODIS Products Classification Using Deep Learning 63
X2 w1
X1 w2
f S
n
s= f Wi X i
Input Hidden layers Output i=1
a) b)
Fig. 1. BASIC architecture of a multilayer artificial neural network
model that best classifies the information. With this set of data and applying a
learning algorithm called “retro-propagation” algorithm based on the descending
gradient technique, the ANN generates a structure that has the ability to classify
information that is not in the original data set correctly [10].
2.2 Convolutional Neural Network
The Convolutional Neural Networks or CNN, traditionally have been used almost
specifically in the classification of images (Fig. 2). CNN has been used success-
fully in multiple areas such as computer science, medicine, economics, music,
painting, and science, to mention just a few [5–8,11]. In CNN, neurons corre-
spond to receptive fields in a manner similar to neurons in the primary visual
cortex of the biological brain. The CNN is composed of a set of filters of one
or more dimensions used to highlight specific features of the information you
want to classify, at the end of each filter layer, a function that is responsible for
performing a non-linear mapping is applied. Unlike the ANNs, during the CNN
training process a set of specialized filters is obtained instead of wi type weights.
The main operation within CNN is called convolution. The convolution allows
highlighting characteristics or specific aspects of the image and it is precisely
this ability to extract features that have made the CNN widely used successfully
in subjects related to computer vision and classification of images.
2.3 Construction of the Training Data Set

As with all supervised learning algorithms, an appropriately labeled data set is
required to design, train an algorithm capable of classifying information appro-
priately. This data set is used during the training phase of the RN, additionally,
it is necessary to have a set of data for the verification and validation of the same
neuronal network. For the present work, a set of MODIS data products was gen-
erated consisting of 253,500 images of 28 × 28 pixels, which were extracted from
more than 17 years of historical data from products. The study region where
Input C
Fig. 2. General structure of a Convolutional Neuronal Network (CNN)
these images were extracted is located within the Caribbean Sea of Mexico and
part of Central America. The images were obtained by applying a window with
25% overlap between images to preserve the spatial relationships between them
(see Fig. 3). The data set is composed of four classes of MODIS products: chloro-
phyll, sea surface temperature, fluorescence and the photosynthetically available
instantaneous radiation (IPAR).
Fig. 3. Study region to create the data set
3 CNN for the Classification of MODIS Products

In Fig. 4, the convolutional neuronal network designed for the present is shown.
The proposed CNN is composed of two parts; the block of convolutional layers
and the block containing a classifier based on a multilayer perceptron. The block
of convolutional layers is composed of two sets of filters, by one hand, the first
of which consists of 32 filters of 3 × 3 pixels, on the other hand, the second block
is composed of 64 filters of the same size. At the output of each set of filters,
the rectified linear unit activation function (i.e. RELU) is used, finally, in order
to extract the representative information of the last set of filters, in the last
section of the convolutional block a two-dimensional “Max Pooling” operation,
was used. The classifying block (multilayer perceptron) is composed of an input
layer formed by 128 neurons with a RELU-like activation function and an output
layer of 4 neurons with a softmax-type activation function.
Dense 128
Softmax
Dense 4
Relu
Relu
Input
Relu
3x3 3x3 2x2
MP2D
32 64
Convolutional layers Multilayer perceptron
Fig. 4. Convolutional Neural Network for classification of MODIS products
For the implementation of the proposed neural network, the Python 3.7.0
programming language and the library Keras 2.2.4 with TensorFlow 1.10.0 as a
backend were used. TensorFlow is an Open Source library developed by Google’s
Brain Team, it is used for the numerical calculation using “flow graphing pro-
gramming”. The nodes in the graph represent mathematical operations while
the connections within the graph represent multidimensional data sets known as
“Tensors”. Tensor Flow contains a set of automatic learning algorithms as well
as other tools that make it ideal for the development of new methods. Keras is a
library that can be used by Python and provides an abstraction layer that facil-
itates access to TensorFlow to create deep learning models. The convolutional
neural network was trained and tested using a Lenovo workstation with an 8-core
Xeon EP processor with 64 GB of RAM and an Nvidia Quadro k5000 GPU with
1536 parallel processing CUDA cores using the Ubuntu 18.04 operating system.
4 Results
To carry out the training and testing of the proposed neural network, the k-fold
procedure with k = 5 was used as a cross-validation method. The cross-validation
process is used to estimate the ability of an CNN to classify information that
has never been presented to the model (generalization), that is, a limited sample
of information is used to estimate how the Model performs in general when used
to make predictions about data that was not used during model training. K-fold
is one of the most popular cross-validation methods because it results in a less
biased or less optimistic estimate of the model’s capacity than other methods,
such as a simple trial-and-error division. During the training of the proposed
network, 80% of data was used to design (training, and testing) the CNN while
the rest 20% was used for the validation. As it can be seen in Fig. 5, both the
data set and the validation data set are balanced, that is, the total number
200,000
150,000
chlor
ipar
sst
100,000
50,000
chlor
ipar
sst
Training Testing
Fig. 5. Histogram of training and test data sets
of elements for the class used for each of the cases is similar. The purpose of
avoiding biases when generating the model.
The training process was carried out for 20 epochs or iterations, in each
iteration blocks (batch size) of 100 images were used. The optimization method
used during the training phase was Adam’s algorithm. The Adam optimization
algorithm (Addictive moment estimation) was presented by Kingma Diederik
in 2015 [12] and this is an extension of the stochastic gradient method, this
algorithm has been adopted in the field of deep learning for its profound impact
on the applications of computer vision and natural language processing.
Figure 6 shows the convergence graph during the training process, the black
dotted line shows the behavior of the neural network during the training phase, it
is evident, the network is able to obtain a good model of classification relatively
fast. As you can see, at the epoch 10 of the training phase, it is clear that
the neural network has learned to distinguish between different classes with an
accuracy close to 100%. The continuous line of green color shows the validation
of the neural network, that is to say, the capacity of the network to classify data
that has never seen previously in other words, the generalization capability of
the network. Thereby it is clear the overfitting is not present as it can be seen in
the plot result. Finally derived from the behavior and trend of the data during
the training process, it can be clearly distinguished that the network has not
memorized the data, instead, it has generated a robust enough model for the
successful classification of MODIS satellite products.
In Table 1, a summary with the results obtained is shown. As can be seen, the
percentage of successful classification by the neural network was 99%. During
the cross-validation phase, it was found that the architecture is able to classify
up to 99.97% of the information correctly. Also in the same table, it can be seen
100 %
90 %
Accuracy
80 %
70 %
60 %
Trend with training data

50 % Trend with test data
2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0

Epochs
Fig. 6. Convergence plot during the training phase (Color figure online)
that the measures of central tendency and dispersion are similar between the
training and test data sets, which implies that the results, at least from this
classical point of view, they have the same tendency and statistical dispersion.
Table 1. Statistical data of the training process and data sets
Statistical measure Value

Successful classification percentage 99.0
Best percentage of successful classification 99.973831
Training data set size 200,000
Size of the test data set 53,500
Average of the training data set −0.09725764
Average of the test data set −0.09741989
Standard deviation of the training data set 0.78812724
Standard deviation of the test data set 0.78970873
5 Discussion
The classification of images using neural networks and deep learning is not new,
In general, the classification of images is done using three bands mainly (red,
green and blue) however, depending on the complexity of the data set, neural
networks, even convolutional networks, are not able to achieve optimal results. In
many cases, for the correct classification of information, the development of new
specialized architectures is required in such a way as to increase the accuracy of

the classification.
The satellite images, depending on the sensor and platform, are composed
of at least 8 bands (Landsat), in the case of the MODIS sensor, the number of
bands increases to 36, which requires, for their classification, an architecture with
optimal parameters. After a detailed study of the state of the art, no record was
found showing the classification of satellite products such as chlorophyll, surface
water temperature and fluorescence by using modern computation techniques,
so the use of Deep Learning as a classification tool is in itself a contribution to
the state of the art. A drawback of the lack of algorithms in the literature, is
the lack of contenders in order to compare the results obtained with the state of
the art.
The proper selection of the operating parameters of a convolutional network
even of a multilayer neural network is not trivial, even today, it is still an open
topic where in many occasions it becomes more an art than a science. Finally,
the development of libraries such as TensorFlow or Keras allow researchers to
concentrate on solving the problem rather than on the operational implemen-
tation of this, that is why (in addition to the increase in computing power and
storage space) that in recent dates we have observed an exponential increase in
the applications of Artificial Intelligence.
In this work, the appropriate parameters for a convolutional architecture
capable of classifying satellite products derived from the 36 MODIS bands were
developed and obtained. Achieving an accuracy of more than 98% the archi-
tecture shown is simple, however, the selection of the parameters has not been
simple, these have been determined experimentally in order to find the best
network performance.
Another important contribution of this work is the creation of an image
bank, which was generated from the historical collection of more than 17 years of
images obtained by the ERIS-Chetumal Satellite Information Reception Station,
This image bank has information for the entire Mexican Caribbean and it has a
total of 253,500 images, which can be used for other algorithms of classification,
clustering and data processing.
6 Conclusions and Final Comments

The area of remote satellite detection has a great impact in the face of support
in emerging problems and with a great social impact, such as the detection of
sargassum, the study of soil change, the environmental impact in large-scale
works, like the Mayan train among other.
The remote perception without a doubt, still has many challenges ahead such
as for example the automatic classification of information. The use of algorithms
based on artificial intelligence such as CNN presented in this work represent an
alternative to be able to face this type of challenges. As shown in the results,
with the use of convolutional neural networks, encouraging results are obtained
for the automatic classification of MODIS satellite data.
In the present work, a convolutional neural network was presented for the
automatic classification of MODIS satellite products. After the careful elabora-
tion of a data bank with more than 250,000 images of MODIS products divided
and correctly labeled in four classes for a wide region of the Caribbean Sea of
Mexico, a CNN capable of automatically classifying these images was designed.
As a result of using the k-fold cross-validation method, a CNN with a classi-
fication capacity of 99% accuracy was obtained. The application of measures
of central tendency and dispersion showed that both the training data set and
the validation data are statistically consistent, which, together with the results
obtained, helps to conclude that the proposed network is capable of generalizing
with a high degree of confidence.
7 Future Work
As part of the future work, the authors are currently working on the design of
a new architecture of Neural Network for the automatic classification of more
than 20 MODIS products. With a database of more than 1,019,200,000 data
and more than 1,300,000 images for the Caribbean Sea, new storage, access and
information manipulation challenges are being faced, which implies, among other
things, the design of new algorithms as well as the use of from other disciplines
such as data science, Big Data, parallel programming using graphical processing
units clusters and specialized hardware units for the processing of tensors like
Google TPU technology.
References
1. Russell, S.J., Norvig, P.: Artificial Intelligence: A Modern Approach. Pearson Edu-
cation Limited, Kuala Lumpur (2016)
2. Jensen, J.R., Lulla, K.: Introductory Digital Image Processing: A Remote Sensing
Perspective. Prentice Hall, Upper Saddle River (1987)
3. Arellano-Verdejo, J., Lazcano-Hernandez, H.E., Cabanillas-Terán, N.: ERISNet:
deep neural network for Sargassum detection along the coastline of the Mexican
Caribbean. PeerJ 7, e6842 (2019). https://doi.org/10.7717/peerj.6842
4. Lazcano-Hernandez, H.E., Arellano-Verdejo, J., Hernandez-Arana, H.A., Alvarado-
Barrientos, M.S.: Spatio-temporal assessment of “Chlorophyll a” in Banco Chin-
chorro using remote sensing. Res. Comput. Sci. 147(12), 213–223 (2018)
5. Cheng, G., Yang, C., Yao, X., Guo, L., Han, J.: When deep learning meets met-
ric learning: remote sensing image scene classification via learning discriminative
CNNs. IEEE Trans. Geosci. Remote Sens. 56(5), 2811–2821 (2018)
6. Rajkomar, A., et al.: Scalable and accurate deep learning with electronic health
records. NPJ Digit. Med. 1(1), 18 (2018)
7. Poplin, R., et al.: Prediction of cardiovascular risk factors from retinal fundus
photographs via deep learning. Nat. Biomed. Eng. 2(3), 158 (2018)
8. Han, J., Zhang, D., Cheng, G., Liu, N., Xu, D.: Advanced deep-learning techniques
for salient and category-specific object detection: a survey. IEEE Signal Process.
Mag. 35(1), 84–100 (2018)
9. Haykin, S.S., Elektroingenieur, K., Haykin, S.S.: Neural Networks and Learning
Machines, vol. 3. Pearson Education, Upper Saddle River (2009)
10. Michie, D., Spiegelhalter, D.J., Taylor, C.C.: Machine Learning. Neural and Sta-
tistical Classification, 13 (1994)
11. LeCun, Y., Kavukcuoglu, K., Farabet, C.: Convolutional networks and applications
in vision. In: Proceedings of 2010 IEEE International Symposium on Circuits and
Systems, pp. 253–256. IEEE (2010)
12. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization. arXiv preprint
arXiv:1412.6980 (2014)
A Dynamic System Model Using Agile
Method Techniques Applied to Geomatics
René Rodrı́guez Zamora1(B) , Álvaro Peraza Garzón1 ,

and Iliana Amabely Silva Hernández2
1
Universidad Autónoma de Sinaloa,
Av Ejercito Mexicano 1166, Fracc. Tellerı́a, 82140 Mazatlán, Sinaloa, Mexico
rene.rodriguez@info.uas.edu.mx, aperaza@gmail.com
2
Universidad Politécnica de Sinaloa, Carretera Municipal Libre Higueras KM 3,
Colonia Genaro Estrada, 82199 Mazatlán, Sinaloa, Mexico
isilva@upsin.edu.mx
Abstract. In this paper we propose the use of a model that represents

a territory across a dynamic system. As part of the process in the defini-
tion of the model we used techniques adaptive software engineering, par-
ticularly construction tools associated with agile methods. The dynamic
system modeling serves to simulate territorial scenarios and analyze their
characteristics using geographical information systems. The approach is
to design user stories as artifacts that can be used for example by decision
makers, urban planners or researchers to more easily observe and read
territorial scenarios that can be to present in the evolution of land-use
changes in a geographical area of interest. This represents an important
advantage because it reduces reading complexity and facilitates analy-
sis because combining agile methods with a simulation model provide a
macroscopic and simplified view of a scenario of nature complex.
Keywords: Dynamic systems · Geomatics · GIS · Agile methods
1 Introduction
The vertiginous advance of the digital technology has propitiated the whole
transformation in the development of the activities that the persons realize daily.
The evolution in the land use of the territory is a fundamental element in our
society, since it manifests different variables that affect our daily life, for example,
accessibility to different points of interest within the city, slopes of the land, etc.
This evolution has gained interest mainly fueled by the different environmental
problems especially those in urban areas [1–3].
Thanks to the advances in the computing field and the development of impor-
tant analytical tools such as Geographic Information Systems (GIS) or simula-
tion models, the study of the changes taking place in metropolitan areas has been
promoted [4]. The analysis of the environmental alterations that result from these

https://doi.org/10.1007/978-3-030-33229-7_7
72 R. Rodrı́guez Zamora et al.
changes and the development of new planning instruments, has caused that dif-
ferent disciplines, specifically the Artificial Intelligence (AI), approaches from a
computer and mathematical point of view to give alternative solutions to this
problem [5,6]. Numerous modeling tools have emerged in recent years. In the
case of urban growth, the models based on cellular automata (CA) are the most
widely used [7,8].
In this paper we present a discrete dynamic system represented by a cellular
automaton that can be used to simulate a geographical area. We test the model
by doing experiments from a digitalized map and generate a corresponding map
from a simulation. Finally, as an additional element to the model definition we
use user stories and scenario description as agile method techniques for commu-
nicating important aspects that characterize the model.
2 Dynamic Systems
The dynamic systems are systems which internal parameters (variables of the
state) follow a series of temporary rules. There are called systems because they
are described by a set of equations (system) and dynamic because its parameters
change with regard some variable that generally is the time. The study of the
dynamic systems can be divided into 3 subdisciplines:
– Applied dynamics: Modeling processes using state equations that relate past
states to future states.
– Dynamics Mathematics: Focuses on the qualitative analysis of the dynamic
model.
– Experimental Dynamics: Laboratory experiments, computer simulations of
dynamic models.
The dynamic systems can split into two big classes: those in which the
time changes continuously and in which the time passes discreetly. Dynamic
continuous-time systems are expressed with differential equations; these can be
ordinary differential equations, differential equations in derivative partial, and
differential equations with delays. On the other hand if the time is discreet the
systems are described by means of difference equations. In the following figures
some examples of dynamic systems can be observed [9,10].
The cellular automata were conceived at the beginning of the 1950’s by John
Von Neumann, a mathematical brilliant interested in investigating about the
required complexity so that a device could be auto-reproductive and the organi-
zation of this one to have a correct operation when repairing itself with part that
could have and operation badly [11]. The results were not let hope, all a current
of investigation has been developed with strong interrelation in fields like frac-
tals and dynamic systems, as well as applications in: parallel cellular computing,
simulation of dynamic systems and recognition of patterns, and simulation of
socio-economical models, for instance (Figs. 1 and 2).
In general therms, we can say that a cellular automata is defined and operates
based on the following elements [11]:
A Dynamic System Model Using Agile Methods Applied to Geomatics 73
Fig. 1. New product adoption diagram.
Fig. 2. Self-powered dynamic system.

Fig. 3. Sales model.
– States, or attributes, e.g., “on” or “off”, 0 or 1.

– Containers that store those attributes, e.g., a location in and array, a sector
in a computer’s memory.
– Behaviors associated with those states; formulated as conditional transition
rules (IF, THEN, ELSE) that govern how and when automata react to infor-
mation that is input to them.
– Neighborhoods from which the automata draw input, e.g., neighboring loca-
tions in an array, databases containing information.
– Temporal domain, through which the evolution of the automata proceeds,
e.g., real-time, program cycles.
We can define cellular automata so that it operates in one, two or more

dimensions. Nevertheless, the complexity to analyze them or to implement sim-
ulations of these in a computer is exponentially increased with the number of
dimensions and states including in this definition Fig. 3.
Cellular automata are also extremely useful for modeling disease transmission
[12]. By treating each cell as an individual and defining a few key constants we
can model how much damage a given disease can be expected to do. The four
key quantities are the infectivity (the rate at which the disease spreads), latency,
(how long before it becomes active in the host), duration (how long it remains
in the host and mortality (the percentage of people the disease kills).
Patterns show up often in nature, from the coloration of fur on a leopard to
the shape a fern takes as it grows. The way in which these patterns for is poorly
understood. However, Wolfram [13] has theorized that they are generated in
a manner similar to cellular automata. In this model, each pigmentation cell
behaves according to the state each of its immediate neighbors has taken on,
much same way a cellular automata works. Another example of this is in the
way mollusk shells are generated. Since the shell is extruded one cell layer at
a time (much like your fingernails), the complex color patterns may be acting
as a one-dimensional cellular automata. Examples of mollusk shells can be seen

below. Below you can see an example of one such simulation [14] (Fig. 4).
Fig. 4. Some cellular automata.
3 Agile Methods
There are methodologies based on classical paradigms within the scope of soft-
ware development guided by models used at the time by software engineers like
linear models represented in cascade, and the evolutionary, iterative and incre-
mental represented in spiral and prototypes.
One of the most widely used standard methodologies is the Unified Process
(UP) for different application areas and different classes of organizations [15].
There are three characteristics that distinguish the UP from the rest of the
methodologies: is centered in the architecture, is iterative and incremental, and is
oriented and directed by use case. Although these characteristics distinguish UP,
do not leave aside the advantages offered by other methodologies, that is, they
take advantage of their strengths in the development of stages and phases that
are already well established in the field of software engineering, but increasing
the flexibility and giving the first guidelines for which are known as agile methods
[16].
The agile methods are part of a set of methodologies also known like light
which use similar practices based on the results, the persons and its interaction.
Exists a diversity of methodologies as the extreme programming (XP) Crys-
tal Clear, Feature Driven Development (FDD), Adaptive Software Development
(ASD) and Scrum. All these methodologies adopt the one that is known as an
agile manifesto, which values:
– To individuals and their interaction, over processes and tools.

– The software that works over the exhaustive documentation.
– The collaboration with the client, above the contractual negotiation.
– The answer to the change over the follow-up of a plan.
These four postulates are implemented using agile principles such as: satisfying
the client through an early and continuous delivery of the software, the fact that
the requirements are changing, the elements of an organization are positively
assumed. They must work together on a daily basis during the development
of the project, it is established that the best architectures, requirements and
designs emerge from teams that self-organize.
Agile methodologies seek at all times teamwork, multidisciplinary and col-
laborative, which allows to carry out projects with immediacy and flexibility,
adapted to changes, demanding at any time modify the mentality, a cultural
change based on respect, responsibility, values, competencies and skills of the
person.
An agile methodology implements a series of practices (values, aptitudes and
skills) short and repetable, that is to say, it is a question of iterating repeatedly
to be adapting changeable requisites inside the frame of the scope of the project.
It seeks to get ready entregables often, with preference to the period of time as
short as possible, to refine and to converge to an acceptable solution.
In addition to the model incremental, some other differential aspects of agile
methodologies are, without doubt, the following:
– Equipment self-governed, self-organized and multi-functional. It is not neces-

sary to have a team leader, since the team itself is able to self-regulate.
– The team regularly reflects on how improving the efficiency, and fits to obtain
it.
– The motivation of the team members is absolutely essential.
– Prioritizes face-to-face communications against the excessive documentation.
– Accepts without problems changing requirements that, in fact, are a funda-
mental part of its rationale.
– There are realized deliveries of the functional product with a frequency of
between 1 to 4 weeks. It is the main progress measurement.
– Sustainable development with a constant rhythm.
– Search for technical excellence and the best possible design.
– One looks for the simplicity, to maximize the quantity of work that is not
necessary to do, to do as well as possible the rest.
Within the set of agile methodologies is located SCRUM [17,18], which provides
a number of tools and roles for, iteratively, be able to see the progress and results
of a project.
A way to use user stories [19,20] to define in this case part of the dynamic
system model is through the technique BDD (Behavior Driven Development) in
which integrates the history a scenario description of behavior.
Fig. 5. Stories Scenarios Options.
4 Model Definition
The fundamental idea in CA Models is that the state of a cell at any given
time depends on the state of the cells within its neighborhood in the previous
time step, based on a set of transition rules [21]. The CA model used in this
investigation is the one proposed by R. White (see Eq. 2), is a constrained cellular
automata for high-resolution modelling of urban land-use dynamics [22,23]. The
CA model is oriented towards the representation of the attributes of a given
geographic region in a two-dimensional lattice, raster maps provides these data
format to the CA.
A raster map can be represented formally by an array of real values. This
matrix is represented as A = aij of order mxn such that 0 ≤ i ≤ m, 0 ≤ j ≤ n
where each element A = [aij ] ∈ R.
A neighborhood filter matrix 1 is required to analyze each element A = aij ,
this neighborhood is formally represented B = bii of order nxn such that 0 ≤
i ≤ n where each element B = [bii ] ∈ Z.
⎡ ⎤
bi−1j−1 bi−1j bi−1j+1
B = ⎣ bij−1 bij bij+1 ⎦ (1)
bi+1j−1 bi+1j bi+1j+1
The neighborhood filter is used to calculate the transition potential from state h
to j for each element A = [aij ]. The calculation methodology is detailed below:
⎛ ⎞

Phj = vsj aj ⎝1 + mkd Iid ⎠ + Hj (2)
k,i,d
where
Phj : is the transition potential of state h to state j.
x
v: stochastic perturbation term. v = 1 + [−ln(random)] . 0 < random < 1,
and x allows you to adjust the size of the disturbance.
sj represents the suitability of the state of the cell.
aj : euclidean distance from the cell to the nearest road.
mkd : calibration matrix, contains the weights of each cell as a function of its
state k and distance d.

1, i = k
I id =
0, i = k
where i is the index of the cell in the current neighborhood, k.
The transition potential Phj of each cell Aij is calculated only if the suitability of
the objective state sj > 0. That is, for each cell (pixel) in the map, its transition
potential will be calculated except for those in which its suitability is equal to
zero. For the neighborhood calculation, the calibration matrix mkd gives each
neighbor cell bii a weight based on its state and distance (subscript d in 2)
concerning the analyzed cell aij . The nearby neighbor cells will generally have a
higher weight, positive values are taken for an attractive effect and negative for
repulsive effect, these values tend to decrease as the distance increases between
the analyzed cell and its neighbor, this is called Distance Decay Effect. When
analyzing the neighbor cells, the Iid component helps to filter (multiplying by
1) cells with the same state.
Visually, in Fig. 6(a) we have the urban land use map, in Fig. 6(b), the neigh-
bor is set to 3 × 3 around the analyzed cell. Figure 6(c) calculates the transition
potential of each cell from its current state to a desired state, the higher is
selected. For this case we set as the higher to urban use. Figure 6(d) analyzed
cell change its value to the higher urban use. Figure 6(e) shows observation win-
dow moves to the next cell. An epoch has been completed when the last cell of
the map is calculated. A simulation may require one or more epochs. If we take
into account that this calculation must be done for each pixel of the map, we
find a problem of computational complexity O(2n ), this means, larger size of the
input maps would increase the execution time of the simulation exponentially.
To handle this complexity we used the OpenMP library.
5 Results
From the model of dynamic system using a celular automata we take as param-
eters to define the states and the evolution space a map previously digitalized
by means of a geographic information system (see Fig. 5). On the map are rep-
resented the land uses of an urban region of northwest Mexico. The Fig. 6 rep-
resents the result of the evolution of the celular automata model with the same
Fig. 6. Transition potential process.
land uses for the same period. As can be seen in the evolution of the automaton
there are 10 land uses. These land uses are expressed as user stories incorporating
scenarios defined based on conditions that influence the permanence or change of
state (land use) of the evaluated cell. With the above what we are looking for is
that this can facilitate the characterization and communication with those who
will read such a map, for example, an urban planner. Finally, with this model
it is possible to carry out experiments later for other eras, which will allow to
build territorial scenarios in the future to identify trends in the territorial space
that are taken to obtain the parameters of the model (Figs. 7 and 8).
Fig. 7. Raster map 2004.
Fig. 8. Land-use map generated by CA.
6 Concluding Remark
Urban development and the territorial scenarios that derive from it are topics
of increasing interest. Phenomena such as globalization and migration that pro-
duce among other things, changes in land uses. It is important to look for these
changes in land uses to be sustainable to maintain balance in the various ecosys-

tems that coexist. Thus, we present in this paper some maps obtained from a
dynamic system model represented by a cellular automaton.
We use cellular automata due to its ability to represent complex behaviors
from simple rules in its definition. Although reading and analyzing results is
not a simple thing, mainly if they need to be transmitted to specialists from
other areas. For instance urban planners or anyone who can use this information
to support decision-making or in the development of a plan or project related
to urban development. In this sense, we propose as an innovative additional
contribution the integration of artifacts used in agile methods. Particularly we
propose used user stories as part of a narrative that allows to describe and read
in a simple way the scenarios possible in land-use changes. Seen in the latter as
user stories that clearly illustrate trends for various territorial scenarios.
References
1. Plata-Rocha, W., Gomez-Delgado, M., Bosque-Sendra, J.: Simulating urban
growth scenarios using GIS and multicriteria analysis techniques: a case study
of the Madrid region, Spain. Environ. Plan. B Plan. Des. 38, 1012–1031 (2011)
2. Li, S., Juhász-Horváth, L., Pedde, S., Pintér, L., Rounsevell, M.D., Harrison, P.A.:
Integrated modelling of urban spatial development under uncertain climate futures:
a case study in Hungary. Environ. Model Softw. 96, 251–264 (2017)
3. Gounaridis, D., Chorianopoulos, I., Koukoulas, S.: Exploring prospective urban
growth trends under different economic outlooks and land-use planning scenarios:
the case of Athens. Appl. Geogr. 90, 134–144 (2018)
4. Ralha, C.G., Abreu, C.G., Coelho, C.G.C., Zaghetto, A., Macchiavello, B.,
Machado, R.B.: A multi-agent model system for land-use change simulation. Env-
iron. Model Softw. 42, 30–46 (2013)
5. Yalew, S., et al.: Land-use change modelling in the Upper Blue Nile Basin. Envi-
ronments 3(3), 21 (2016)
6. Pontius, R.: Criteria to confirm models that simulate deforestation and carbon
disturbance. Land 7(3), 105 (2018)
7. Feng, Y., Qi, Y.: Modeling patterns of land use in Chinese cities using an integrated
cellular automata model. ISPRS Int. J. Geo-Inf. 7(10), 403 (2018)
8. Aguilera Benavente, F., Plata-Rocha, W., Bosque Sendra, J.: Diseño y simulación
de escenarios de demanda de suelo urbano en ámbitos metropolitanos. Rev. Int.
Sostenibilidad, Tecnol. y Humanismo 57–80 (2009)
9. Khoshnoud, F., Dell, D.J., Chen, Y.K., Calay, R.K., de Silva, C.W., Dwhadi,
H.: The concept of self-powered dynamic systems. In: European Conference for
Aerospace Sciences (EUCASS), Munich, Germany, 1–5 July 2013. EUCASS (2013)
10. Borshchev, A.: The big book of simulation modeling: multimethod modeling with
AnyLogic. AnyLogic North America, Chicago (2013)
11. Zamora, R.R., Vergara, S.V.C.: Using de Bruijn diagrams to analyze 1d cellular
automata traffic models. In: Sloot, P.M.A., Chopard, B., Hoekstra, A.G. (eds.)
ACRI 2004. LNCS, vol. 3305, pp. 306–315. Springer, Heidelberg (2004). https://
doi.org/10.1007/978-3-540-30479-1 32
12. Beauchemin, C., Samuel, J., Tuszynski, J.: A simple cellular automaton model for
influenza A viral infections. J. Theor. Biol. 232(2), 223–234 (2005)
13. Wolfram, S.: A New Kind of Science, vol. 5. Wolfram Media, Champaign (2002)
14. Brock, Z., Karst, N., Siripong, M.: Bio-inspired computing (2005). ca.olin.edu
15. Jacobson, I., Booch, G., Rumbaugh, J.: The Unified Software Development Process.
Addison-Wesley, Boston (2000)
16. Priolo, S.: Métodos ágiles. USERSHOP (2009)
17. Abrahamsson, P., Salo, O., Ronkainen, J., Warsta, J.: Agile software development
methods: review and analysis. arXiv preprint arXiv:1709.08439 (2017)
18. Cockburn, A.: Agile Software Development. Addison-Wesley Professional, Boston
(2001)
19. Patton, J.: User Story Mapping: Discover the Whole Story, Build the Right Prod-
uct. O’Reilly Media Inc., Cambridge (2014)
20. Trkman, M., Mendling, J., Krisper, M.: Using business process models to better
understand the dependencies among user stories. Inf. Softw. Technol. 71, 58–76
(2016)
21. Barredo, J.I., Gómez, D.J.: Towards a set of IPCC SRES urban land-use scenarios:
modelling urban land-use in the Madrid region, pp. 1–16. European Commission -
DG Joint Research Centre Institute for Environment and Sustainability, Depart-
ment of Geography, University of Alcalá (2000)
22. White, R., Engelen, G.: High-resolution integrated modelling of the spatial dynam-
ics of urban and regional systems. Comput. Environ. Urban Syst. 24, 383–400
(2000)
23. White, R., Engelen, G., Uljee, I.: The use of constrained cellular automata for
high-resolution modelling of urban land-use dynamics. Environ. Plan. B Plan. Des.
24(3), 323–343 (1997)
24. Benavente, F.A., Montes, L.M.V., Lara, J.A.S., Delgado, M.G., Rocha, W.P.: Esce-
narios Y Modelos De Simulación Como Instrumento En La Planificación Territorial
Y Metropolitana. Ser. Geográfica 17, 11–28 (2011)
25. Montgomery, A.W.: Scrum framework effects on software team cohesion, collabora-
tion, and motivation: a social identity approach perspective. Creighton University
(2017)
Telematics and Electronics
Statistical Properties of Vehicle Residence
Times for Fog Computing Applications
David Miguel Santiago1(&), Mario E. Rivero-Angeles2,

Laura I. Garay-Jiménez1, and Izlian Y. Orea-Flores3
1
Instituto Politécnico Nacional SEPI UPIITA, Mexico City, Mexico
dmiguels1800@alumno.ipn.mx, lgaray@ipn.mx
2
Network and Data Science Laboratory,
Instituto Politécnico Nacional (CIC-IPN), Mexico City, Mexico
mrivero@ipn.mx
3
Telematics Department-UPIITA,
Instituto Politécnico Nacional, Mexico City, Mexico
iorea@ipn.mx
Abstract. Currently, new advances in the automotive industry are focused on

implementing autonomous cars, since they are the future to avoid accidents and
protect users. Even if safety is the main goal, many opportunities for techno-
logical developers are possible. Among them, Smart Cities is a major player in
future communication networks. In a smart city context, hundreds or thousands
of sensors will be deployed in many strategic parts of the city, including sensors
in mobile devices, that can provide critical information for the city management
and improve the resident’s livelihood. However, this scenario entails extremely
high volumes of information to be sent to different geographical locations.
Because of this, the use of cellular base stations may be a highly expensive
alternative.
In this work. We propose to take advantage of the use of autonomous cars as
data mules for the efficient recollection of data in smart cities environments.
Specifically, we consider interest points in Luxembourg City, where relevant
data may be generated by sensors in mobile devices or fixed sensors in the city’s
infrastructure. Assuming that, autonomous vehicles know in advance the route
that they are going to follow to reach their destination, sensors can profit the
passage of these vehicles to transmit their data, making short-range, low-cost
transmissions and reducing the implementation cost of these applications. Later,
the vehicles can relay the data on the destination point. To this end, we evaluate
the potential use of this system by obtaining the main statistics variables of the
passage of the vehicles though these interest points in the city. We obtain the
mean, variance and coefficient of variation of the resident times of vehicles to
estimate the potential use of this communication system in Smart Cities.
Keywords: Fog Computing Vehicle networks Statistical characterization

Smart Cities

https://doi.org/10.1007/978-3-030-33229-7_8
86 D. Miguel Santiago et al.
1 Introduction
Smart Cities are envisioned to be a major component of people’s everyday lives in big
cities for the near future [1]. This technology will connect a high number of sensors
placed throughout the city to monitor highly sensitive processes in the city, such as,
water supply, gas or toxic substances leakages, pollution levels, temperature, floods,
urban wildlife (rats, dogs, cats, birds, etc.) monitoring [2], and even traffics jams [3]
and criminal acts. Sensors can be integrated into the urban infrastructure but also,
mobile devices which nowadays have many different types of sensors (gyroscopes,
acceleration, sound, video, heart rate monitoring, light, etc.) [4], can be used to gen-
erate and convey relevant data for the city operation.
For instance, urban wildlife monitoring apps can send an alert for people in a
specific point in the city, like a public park, to take sound samples searching for birds of
interest. By recognizing the individual sound characteristics of birds in the environ-
ment. In this application, only users and sensors placed in a certain region are asked to
send their data, while sensors and mobile devices outside this region are not involved in
the data recollection process. This leads to the concept of Fog Computing [5], where
devices in a certain region, inside the fog, communicate and process information that
does not need to reach the cloud computing [6], i.e., this information is locally gen-
erated, stored and transmitted.
Also, relevant in the context of Smart Cities is the development of autonomous
vehicles that will help us with our goal. For years, driver assistance, automation sys-
tems, and communications have been developed to guide us to the next generation of
mobility [7]. The complexity of these developments tries to connect vehicles with other
vehicles or with a controlled environment out of the car. Autonomous vehicles have the
great advantage that the route can be known in advance. Indeed, before driving begins,
the vehicle can calculate the adequate route based on the traffic conditions, driver’s
personal preferences, accidents in the routes, and weather conditions among many
others.
Some works talk about how to make the roads safer and efficient the vehicular
traffic through communications V2V (Vehicle-to-Vehicle) and V2I (Vehicle-to-
Infrastructure) [8], in the last concept it is necessary to use the RSUs (Road-Side
Units), i.e., devices that provide connectivity with vehicles in the range [9], this work
has a similar concept in comparison with this paper but there are differences. We
propose a fixed or mobile sensor to analyze in a determinate area, as Fog Computing, if
a car can or cannot receive data to be used as a data mule without the necessity to have
an RSU because the information can be received by a pedestrian or another mobile
device.
On the other hand, there are advances as VANET’s (Vehicular Ad hoc NETwork)
[10] where a vehicle is considered as a communication unit and vehicles can trade
information without a network or Fog Computing, this work is different than our
propose. We make an emphasis on participation and collaboration between pedestrians
and automobiles.
In this work, we make a study of the communication between vehicles and a sensor
placed in specific parts of the city. Specifically, we consider a system where a sensor,
Statistical Properties of Vehicle Residence Times 87
fixed or mobile, generate information in strategic parts of the city and take advantage of
the passage of autonomous cars. Since the routes of these vehicles are well defined, the
sensor devices can determine if the passing vehicle can be used as a data mule to
convey their data to another part of the city. Although, in this paper, the main goal is to
analyze if the interaction between vehicles and a sensor, fixed or mobile, can be able
viable or not.
To characterize these residence times, we use real vehicle trajectories from a well-
known database on SUMO (Simulation of Urban MObility) [11] simulator about the
traffic of vehicles in different parts of the city to analyze the behavior of vehicles in
different traffic conditions.
SUMO [12] is a software that can provide a vehicular scenario of a real-world
working with JOSM (Java OpenStreetMap) that provides the updated map of the city.
For this analysis, we work with an article published by the University of Luxem-
bourg [13] called “Luxembourg sumo traffic (lust) scenario: 24 h of mobility for
vehicular networking research.” where it is possible to obtain a simulation scenario in a
period of 24 h of traffic from Luxembourg. This work was compared with the Typical
Traffic option in Google Maps and showed similar traffic with the real traffic in
Luxembourg.
Then, we calculate the times that an automobile is in a street, waiting in a traffic
light or moving with a specific speed. The position and the behavior of a vehicle are
important parameters used to calculate the residence times of vehicles in a fog com-
puting application. Luxembourg is a typical European city, and we assume that the
results presented in this paper can be extended to other cities with similar conditions.
The realistic scenario was measured in a common day, during the morning the
traffic is high density, in the afternoon is medium density and, in the night, the density
is lower than other times of the day. In this scenario, there are vehicles, buses, and
pedestrians that have different behaviors.
The rest of this paper is organized as follows: Sect. 2 presents the main assump-
tions and system set-up were the residence times where measured. Then, in Sect. 3 we
explain how the characteristics of the residence times were obtained. Finally, Sect. 4
presents relevant results.
2 System Model
The vehicular analysis consists in considering 24 h of traffic observation in Luxem-

bourg city. We focused on low density (during the night). In this real vehicle trajec-
tories, 295,979 vehicles interacted in an area of 155.95 km2. As such, it is
computationally difficult to analyze all possible interactions between cars during all
24 h. Hence, we selected periods of ten minutes for each kind of traffic density and we
identified five different places considered as tourists spots where a high vehicle con-
centration is expected. We selected these places based on recommendations of touristic
attractions. In Fig. 1, we present the SUMO vehicle trajectories and we compare it to
Google maps to identify both the touristic spots where many sensors (fixed and mobile)
can gather relevant information and the traffic conditions to use vehicles as data mules.
Fig. 1. Comparison between map in SUMO and Google maps.
We can observe that both images are similar, but some routes do not exist in the
SUMO map. Then, based on both maps, we selected specific places to analyze
vehicular concurrence. These points represent parts of a smart city scenario where
information can be generated. Hence, all sensors (fixed and mobile devices) inside the
range of interest will form the Fog Computing network, i.e., data is sensed, stored,
processed and conveyed to passing vehicles to then transmit it to other parts of the city.
Also, note that this information is not relevant to other parts of the city. As such, this
information does not have to reach the cloud, effectively reducing the traffic and
resource utilization in the cloud. For instance, a WiFi-based network has a range of
about 100 m and Bluetooth about 10 m. Both these networks are typically enabled in
any commercial smartphone and sensor nodes. We also considered ranges of 50 m and
100 m for completeness purposes, then we make a radius of 25 m and 50 m around the
point to simulate a network. In Table 1, we showed the selected places.
Table 1. Places where residence vehicular were analyzed.

Place Direction Range
Schlassbréck bridge Montée de Clausen Av. & Rue Sisthéne Weis 50 m,
100 m
Justice Palace of Franklin Delano Rooselvet Roundabout & René 50 m,
Luxembourg Konen Tunnel 100 m
Art Gallery Schorten Emile Reuter Av. & Royal Roundabout 50 m,
100 m
Sofitel Luxembourg Le Boulevard d’Avranches & Rue du Laboratoire 50 m,
Grand Hotel 100 m
International School of Rue de Bouillon Roundabout 50 m,
Luxembourg 100 m
The first place is a bridge and is an intersection between Montée de Clausen street
and under Rue Sosthéne Weis street, this location is near Casemates du Bock that is an
antique architecture to defend the city called Schlassbréck bridge. As seen in Fig. 2, the
radius is in red and, this circle simulates the network, i.e., all vehicles that enter on this
circle is considered as potential data mule to convey data to other parts of the city.
Fig. 2. The first place to analyze in SUMO map and Google map.
The second point of interest is shown in Fig. 3 and is located near Justice Palace,
this place is between Franklin Delano Roosevelt boulevard and René Konen tunnel.
Fig. 3. The second place to analyze in SUMO map and Google map.
The third point of interest is depicted in Fig. 4 and is near the Art Gallery Schorten
and is located between Emile Reuter avenue and Royal boulevard.
The fourth point of interest is shown in Fig. 5 is near a hotel called Sofitel
Luxembourg Le Grand Hotel located in d’Avranches boulevard and Rue du Labora-
toire street.
Fig. 4. The third place to analyze in SUMO map and Google map.
Fig. 5. The fourth place to analyze in SUMO map and Google map.
The last place is near the International School of Luxembourg Asbl and the location
is in Rue de Bouillon boulevard and it is shown in Fig. 6.
Fig. 6. The last place to analyze in SUMO map and Google map.
3 Distribution of Residence Times
Once the interest points are selected and the different communication ranges are placed
around it, we now can calculate the residence times of all vehicles going through the
circles described in the previous section. We analyze the information making a program
in python because SUMO generates files with XML format. These files have several
data characteristics about the vehicles but the information that we need is the position
and time of each vehicle. The position is given as cartesian coordinates and, as men-
tioned before, we consider periods of 10 min because it was difficult to analyze the
24 h. The 10 min period is equal to 600 s; during this time, we analyze the position of
each vehicle second inside the coverage data.
To find vehicles in a circumference we used the circumference equation given by:
x2 þ y2 ¼ r 2 ð1Þ
Then, we added our coverage range to localize the vehicles inside the interest points
as follows:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
50 ðx OrigxÞ2 þ ðy þ OrigiyÞ2 ð2Þ
or
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
100 ðx OrigxÞ2 þ ðy þ OrigiyÞ2 ð3Þ
Where x and y are the coordinates of each vehicle and Origx and Origy are the interest
points and all vehicles present in a radius equal or minor to 50 or 100 meters are
considered in this analysis.
In this work, we are only interested in studying the statistical characteristics of this
residence time. Then, we calculate the mean, variance and coefficient of variation of
these residence times. The main data that we are going to use is the coefficient of
variation because is a relation between the size of the mean and the standard deviation.
r
CV ¼ ð4Þ
l
These are some properties of CV.

1. It does not have units.
2. Based on these characteristics we can propose a distribution with similar behavior
according to the following cases:
3. In the case that the CV = 1, the possible distribution that can describe the residence
times is the exponential distribution. This distribution describes the time elapsed
between two independent events.
4. In the case that the CV > 1, the possible distribution that can describe the residence
times is the Hyper-exponential distribution.
5. In the case that the CV < 1, the possible distribution that can describe the residence
times is the Erlang distribution. This distribution describes the wait time until an
event occurs.
4 Numerical Results
We now present some relevant results of the residence times. In Figs. 7 and 8 we
present the histograms of the residence times in the first point of interest considering
low vehicle density with 50 and 100 m of coverage radio respectively.
Fig. 7. Histogram in Schlassbréck bridge with 50 m radius.
Fig. 8. Histogram in Schlassbréck bridge with 100 m radius.
For the second point of interest, we present the histograms in Figs. 9 and 10.
For the third point of interest, we present the histograms in Figs. 11 and 12.
Fig. 9. Histogram in Justice Palace with 50 m radius.
Fig. 10. Histogram in Justice Palace with 100 m radius.
Fig. 11. Histogram in Art Gallery Schorten with 50 m radius.

Fig. 12. Histogram in Art Gallery Schorten with 100 m radius.
For the fourth point of interest, we present the histograms in Figs. 13 and 14.
Fig. 13. Histogram in Sofitel Luxembourg Le Grand Hotel with 50 m radius.

Fig. 14. Histogram in Sofitel Luxembourg Le Grand Hotel with 100 m radius.
Finally, for the fifth point of interest, we present the histograms in Figs. 15 and 16.
Fig. 15. Histogram in International School of Luxembourg Asbl with 50 m radius.
In each figure, we show the mean (µ), the variance (r2 ) and the coefficient of
variation (CV). As we mentioned before, the CV is part of our interest since we can
choose a hypothetical distribution that matches the statistical characteristics of the
residence times. Building on this, we can see that the CV of all residence times
measured in this work are lower than 1. Hence, the Erlang distribution can be used to
model the times that the vehicles remain in the interest region.
Fig. 16. Histogram in International School of Luxembourg Asbl with 100 m radius.
5 Conclusion and Future Work
In this work, we characterized the residence time of real traces of vehicles inside an
interest region in Luxembourg City. To this end, we first locate touristic locations that
are potential points where data can be gathered and disseminated for a future smart city
scenario and autonomous vehicles, although if we could select several locations maybe
we would expect similar results.
Another relevant result is that for the scenarios considered, an Erlang distribution is
a good candidate distribution to model the residence times for future Fog Computing
applications in smart city environments, we could link with the real results and test if
these results are real and not a coincidence. We must prove mathematically.
If we could analyze all data, we would expect similar results, because we consider
that vehicles have the same behavior.
These results can be used for future mathematical and simulation works to calculate
the amount of information that can be conveyed to vehicles acting as data mules to
transport information to other parts of the city in a low-cost scenario, it depends on the
future works.
References
1. Albino, V., Berardi, U., Dangelico, R.M.: Smart cities: definitions, dimensions, performance,
and initiatives. J. Urban Technol. 22(1), 3–21 (2015)
2. Khan, Z., Anjum, A., Soomro, K., Tahir, M.A.: Towards cloud based big data analytics for
smart future cities. J. Cloud Comput. 4(1), 2 (2015)
3. Biswas, S.P., Roy, P., Patra, N., Mukherjee, A., Dey, N.: Intelligent traffic monitoring
system. In: Satapathy, S.C., Raju, K.S., Mandal, J.K., Bhateja, V. (eds.) Proceedings of the
Second International Conference on Computer and Communication Technologies. AISC,
vol. 380, pp. 535–545. Springer, New Delhi (2016). https://doi.org/10.1007/978-81-322-
2523-2_52
4. Ong, R., et al.: Traffic jams detection using flock mining. In: Gunopulos, D., Hofmann, T.,
Malerba, D., Vazirgiannis, M. (eds.) ECML PKDD 2011. LNCS (LNAI), vol. 6913,
pp. 650–653. Springer, Heidelberg (2011). https://doi.org/10.1007/978-3-642-23808-6_49
5. Stojmenovic, I., Wen, S.: The fog computing paradigm: scenarios and security issues. In:
2014 Federated Conference on Computer Science and Information Systems, pp. 1–8. IEEE
(2014)
6. Capponi, A., Fiandrino, C., Kliazovich, D., Bouvry, P., Giordano, S.: A cost-effective
distributed framework for data collection in cloud-based mobile crowd sensing architectures.
IEEE Trans. Sustain. Comput. 2(1), 3–16 (2017)
7. Sookhak, M., et al.: Remote data auditing in cloud computing environments: a survey,
taxonomy, and open issues. ACM Comput. Surv. (CSUR) 47(4), 65 (2015)
8. Aissaoui, R., Menouar, H., Dhraief, A., Filali, F., Belghith, A., Abu-Dayya, A.: Advanced
real-time traffic monitoring system based on V2X communications. In: 2014 IEEE
International Conference on Communications (ICC), pp. 2713–2718. IEEE (2014)
9. Richter, A., Friedl, H., Scholz, M.: Beyond OSM–alternative data sources and approaches
enhancing generation of road networks for traffic and driving simulations. In: SUMO2016–
Traffic, Mobility, and Logistics, pp. 23–25 (2016)
10. Ge, X., Li, Z., Li, S.: 5G software defined vehicular networks. IEEE Commun. Mag. 55(7),
87–93 (2017)
11. Kastner, K.H., Pau, P.: Experiences with SUMO in a real-life traffic monitoring system. In:
SUMO 2015–Intermodal Simulation for Intermodal Transport (2015)
12. Liu, W., Wang, X., Zhang, W., Yang, L., Peng, C.: Coordinative simulation with SUMO and
NS3 for vehicular ad hoc networks. In: 2016 22nd Asia-Pacific Conference on Commu-
nications (APCC), pp. 337–341. IEEE (2016)
13. Codeca, L., Frank, R., Engel, T.: Luxembourg sumo traffic (LuST) scenario: 24 hours of
mobility for vehicular networking research. In: 2015 IEEE Vehicular Networking
Conference (VNC), pp. 1–8. IEEE (2015)
Text and Image Transmission and Reception
Using Light from LEDs and a Light Sensor
Sergio Sandoval-Reyes(&)
Centro de Investigación en Computación (CIC), Instituto Politecnico Nacional,

Av Juan de Dios Batiz s/n, Col. NI Vallejo, 07738 Mexico City, Mexico
sersand@cic.ipn.mx
Abstract. Visible Light Communication or VLC uses visible light from light
emitting diodes (LEDs) to transmit information. Using a computing device and
some hardware, the transmission of information is performed driving and
modulating the light emitted by the LEDs. In the receiver side, the information
carried by the modulated light is demodulated through a light sensor, connected
to a similar computing device for the final recovering of the information. In this
article we describe an application based on VLC using OOK (On-Off Keying)
modulation, to transmit and receive a line of text and an image from a Raspberry
Pi computer (using Python as the programming language), two yellow LEDs
and a sensor light.
Keywords: Text and image VLC OOK Raspberry Pi Python
1 Introduction
Visible Light Communication [1, 2] can be used to transmit audio, voice and data using
light from emitter diodes (LEDs), and light detectors at the transmitter and receiver
ends respectively (Fig. 1). VLC works in the 380 nm to 780 nm optical band which is
visible light [3–5].
To convey information this one has to be encoded, and then the light has to be
modulated and demodulated at the transmitter and receiver sides. There are several
methods to do this, some are briefly discussed in the following. Then, the received
information has to be decoded and processed to recover it fully. The success of this
recovery depends of several factors, among them: (1) The number, shape, and wave-
length of the LEDs employed; (2) The number and type of light detectors (light-sensor,
PIN-diode, reverse-biased LED, etc.) used; (3) The encoding method (RZ, NRZ,
NRZP, etc.); (4) The modulation scheme (OOK, WPM, VWPM, PPM, OFDM, etc.),
and (5) The synchronization and distance between the LEDs and the light detector.
VLC is a technology that requires line-of-sight between both emitter and receiver [6].
In this work we describe an application based on VLC using OOK (On-Off Keying)
modulation, to transmit a line of text and a color image using a Raspberry Pi 3
computer as the data source and sink (to simplify the synchronization problem between
emitter-receiver), Python as the programming language, and several modules (LEDs
and a sensor light) from littleBits, to easy the hardware implementation.

https://doi.org/10.1007/978-3-030-33229-7_9
Text and Image Transmission and Reception Using Light from LEDs 99
Fig. 1. A basic VLC link structure.
This paper is organized as follows: Part 2 presents a summary of works related to

the transmission of text and images using VLC. Section 3 describes the design of our
VLC application. Part 4 describe the experiments realized, the results obtained, and
their analysis. The conclusions and future work are presented in Sect. 5.
2 Related Work
Several research works on VLC technologies to transmit text and images have been
proposed. The most important are described in the following.
2.1 How Based VLC Text and Image Systems Work

Generally, VLC image systems are implemented modulating the intensity of light and a
direct detection scheme with a line-of-sight (LOS) configuration [7]. In the transmitter,
intensity modulation is implemented through the modulation of the transmitted signal
into the instantaneous optical power of the LED by controlling the radiant intensity
with the forward current through the LED (High modulation frequencies are used to
avoid flicker). In the receiver, the transmitted signal is recovered using direct detection
throughout a light sensor to convert the incident optical signal power into a propor-
tional current. The following Fig. 2 shows a general flow chart for VLC text and image
transmission [8].
2.2 VLC Text and Image Transmitter

A typical based VLC text and image transmitter contains a text and image generator (a
PC with Matlab to convert the text and the image into bits), an interface (usually a USB
cable) to send the bits to a microcontroller (for coding and modulation), and taken
through one of its output ports to the LED driver and the LED optics. See Fig. 3 [9].
The modulated signals are used to switch on-an-off the LEDs at desired frequencies
using LED drivers. These drivers rely on transconductance amplifiers to convert
voltage signals into corresponding current signals to excite the LEDs array for com-
munication purposes.
100 S. Sandoval-Reyes
Fig. 2. A flow chart for VLC text and image transmission.
Fig. 3. VLC text and image transmission.
2.2.1 VLC Modulation

There exist different modulation schemes for VLC, the most common are: on-off
keying (OOK), variable pulse-position modulation (VPPM), color shift keying
(CSK) and orthogonal frequency division multiplexing (OFDM) [10]. However, OOK
is the most commonly used modulation scheme in VLC due to its simple implemen-
tation. In this method essentially the LED intensity is changed between two distin-
guishable levels corresponding to the data bits (1 or 0). See Fig. 4. A modified OOK
can also provide dimming. It is achieved by changing the data duty cycle through
pulse-width modulation (PWM), with only 1 bit of information carried per symbol
period.
Fig. 4. On-Off Keying with PWM.
2.3 VLC Text and Image Receiver

A typical optical text and image receiver consists of a photo detector followed by an
amplifier. The photo detector can be a photo transistor, a reverse-biased LED, or a
Light Detect Resistor (LDR). The light captured by the photo transistor which acts as a
sensor, passes the output to the comparator which compares the binary input, and
similarly the original image is recovered using Matlab software in the PC. See Fig. 5
[8, 9].
Fig. 5. VLC text and image reception.
3 Text and Image Tx and Rx Application Using VLC
In this section we develop a VLC application to transmit and recover a line of text first,
and then a color image using two LEDs in the transmitter, and a light sensor in the
receiver.
We begin using in the transmitter as a data source, a Raspberry Pi 3 (RBPi)
computer, and six littleBits bit-modules [12]: power, button, proto, split, and two bright
LEDs. See Fig. 6. The RBPi using a software program written in Python will read a
line of text or an image byte by byte, and will transmit each byte using the SPI MOSI
(Serial Peripheral Interface, Master Output Slave Input) output port (Pin 19), toward the
proto module, OOK signals to drive the two LEDs.
Similarly, in the receiver we will use three littleBits components: power, light
sensor, and another proto bit. See Fig. 7. The light sensor captures the light emitted by
the two LEDs and converts it into a digital signal which is fed to the proto module,
which in turns outputs this signal and with a wire connector, feeds this signal toward
the MISO (Master Input Slave Output) input port (Pin 21) of the RBPi.
Fig. 6. Raspberry Pi 3, and six modules: power, button, proto, split and two LEDS
Fig. 7. Receiver components: light sensor and proto bits.

3.1 Transmitting a Line of Text and an Image via VLC

The program hola_lifi.py written in python transmits a line of text. See Fig. 8.
# hola_lifi.py
# Python code to transmit and recover the line of text “hola Li-Fi” using SPI and VLC
import spidev
spi = spidev.SpiDev()
spi.open(0,0)
spi.max_speed_hz = 4000000
buffer = bytearray()
tx_data = [ord(H”), ord(“o”), ord(“l”), ord(“a”), ord(“ “), ord(“L”), ord(“i”), ord(“-“), ord(“F”),
ord(“i”)]
while True:
# Transmit the text “Hola-Li-Fi”, throught the Raspberry out pin 19 SPI-MOSI,
# and receive it in the array rx_data, throught the Raspberry input pin 21 SPI-MISO
rx_data = spi.xfer(tx_data)
# Receive every byte in rx_data and store them in a buffer to display them in the screen
for byte in rx_data:
buffer.append(chr(byte))
print buffer
Fig. 8. Python code to transmit and recover a line of text via VLC.
Transmitting an image via VLC is very similar to transmit a line of text. Here the
RBPi with a script written in Python, opens a picture file (“lenna.jpeg”) and it reads it
into a bytearray “b”. Then with a “for” loop reads byte by byte of the picture and as
before, it sends them to the SPI MOSI output port 19, using the spi.xfer () directive.
This MOSI output is fed into the input of the proto module (lower middle connector in
Fig. 7). As before, the proto module outputs and splits the OOK signals to drive the
two LEDs; the LEDs light is picked up throughout the light sensor to recover the
image. The Python program to execute the above mentioned is shown in Fig. 9.
In the code above, it is necessary to import the following libraries: “SPI”, “PIL”
(Python Image Library), and “Array”. To drive the LEDs the power module was fed
with a 9-V battery. This was necessary because the outputs of the RBPi are low-voltage
(3.3 V) and low-current (Individual pins must not pull more than 16 mA and the entire
GPIO must not source more than 50 mA), which are no good enough to drive two
bright LEDS [13]. These LEDs are simple yellow LEDS with a wavelength of 550-to-
600 nm, and consume around 16-to-20 mA each, with an aperture angle of about 120°.
See Fig. 10.
3.2 Receiving Text and an Image via VLC

As it was mentioned, the line of text and the image was sent via VLC as LED light.
This light is received through a light sensor module which then sends it back through
out another proto module, to the MISO port 21 of the RBPi. This light sensor not only
receives the OOK light signal but also has a transimpedance amplifier for high speed
operation. The light sensor has two operational modes. In LIGHT mode (switch up in
# lena_spy.py: Python code to read, transmit, and recover a picture via Visible Light
import spidev
from PIL import Image, ImageFilter
import io
from array import array
spi = spidev.SpiDev()
spi.open(0,0)
spi.max_speed_hz = 4000000
b = bytearray()
buffer = bytearray()
try:
# Load an image from Raspberry Pi
with open (“lenna.jpeg”, “rb”) as img:
f = img.read()
b = bytearray(f)
# Sending with SPI and converting the image, byte by byte into visible light
for byte in b:
rx = spi.xfer([byte])
rx_data = rx[0]
buffer.append(rx_data)
# Recovering the bytes array and back into an image
img_rec = Image.open(io.BytesIO(buffer))
img_rec.save(´lena_img_recovered.png´)
img_rec.show()
except:
print “Unable to recover image”
Fig. 9. Python code to read, transmit and recover a picture via VLC.
Fig. 10. LittleBits bright LED.
Fig. 7), as the light shining on the sensor gets brighter, more signal passes through it.
In DARK mode (switch down), the signal increases as it gets darker. Furthermore, the
light sensor has a sensitivity dial or slide dimmer to adjust how much light it takes to
recover the signal, and has a spectral sensitivity range from 500-to-600 nm similar to
the LEDs wavelength. See Fig. 11.
The recovered text and image from the light sensor via the SPI MISO input (pin 21)
is stored into a buffer, saved and displayed, as can be seen from the Python code lines
of Figs. 8 and 9.
Fig. 11. Receiving text and an image via a light sensor.
4 Experiments and Results
For the experiments we use as was mentioned a RaspBerry Pi 3 computer with several
LittleBits components. The whole setup is shown in Fig. 12.
Fig. 12. Setup for the transmission and reception of text and an image using VLC.
Figure 12 shows the whole set up: the Raspberry Pi 3 computer, a HP display,
keyboard and mouse, thus as the LEDs, the sensor light, and the proto modules, fed
with 9 V batteries. Figure 13 shows the recovered text and picture after the execution
of the line commands “sudo python hola_lifi.py” and “sudo python lena_spi.py”.
Figure 13.
Figure 13 also shows that the recovered text and picture was no good enough. That
was due to the presence of noise, mainly: Fluorescent light from ceiling lamps,
Fig. 13. Recovering of text and picture via VLC.
misalignment and distance between the LEDs and the light sensor, and low sensitivity
to light from the light sensor.
4.1 Discussion of Results

After several adjustments to the setup and a few intents, the text and the picture was
finally well recovered. The whole transmission and recovering of a 225 225 pixel
color image weighting 8 Kbytes, took less than a second. See Fig. 14. Naturally, the
noise cannot completely eliminated and increases when the misalignment and the
distance between the LEDs and the light sensor is larger. Also the brightness of the
LEDs influence the performance. That was the main reason for using two LEDs in
parallel to increase the amount of light sent to the light sensor.
Fig. 14. Recovered picture via VLC after some adjustments.
4.2 Efficiency of the Recovered Text and Image via VLC

Due to the fact that the transmission (LEDs) and reception (light sensor) link is made
using Raspberry SPI, two questions arise: (1) What percentage of the image and text
was recovered correctly? and (2) How many times the image or the text was well
recovered, versus the number of times the image or text could not be recovered?
(1) Percentage of the image or text that was recovered correctly: Initially without
any kind of adjustments, the recovery of the image was around 50% as it can be
seen from Fig. 13. Notice from this figure the pixelation of the image in certain
areas. In contrast with the adjustments mentioned before, the image was 100%
recovered as it is shown in Fig. 14. Similar results were obtained for the line of
text.
(2) Number of times the image or text was well recovered: Although the adjust-
ments made to the transmitter-receiver setup, the Tx-Rx link was somewhat
shaky, requiring many tries to obtain a 100% image recovery. The number of
times to get the image well recovered, was 3-out-of-10. In contrast, the recovery
of the line of text required less tries.
4.3 Contributions
This work differs with respect to similar ones reported in [8–11], and [14–18], in the
following. In some of them only a design model is proposed without any proof of
implementation. In others a hardware setup is shown, but again, there is not any proof
of sending and receiving any image. It also required to use two microcontrollers (one
for transmission and one for reception). The main problem of using microcontrollers is
that not all of them run good enough fast and have enough RAM memory to store a
medium to large image sent by the PC. Furthermore, it requires software written in C or
Matlab to reside at both PCs to connect with the microcontrollers and to process the
picture transmission and reception. Some others, do not use visible light at all, but
infrared light. Our work differs from the above in that we only use a Raspberry Pi
computer, visible light from LEDs, littleBits modules, and simple code written in
Python as the programming language.
5 Conclusions
A line of text and an image transmission and reception application using visible light
was developed using a Raspberry Pi 3, two bright LEDs and a light sensor, thus as
OOK modulation, and the Python Image Library. Both the line of text and the picture
in color, were recovered acceptably well although with a small presence of noise. It
should be noted that this noise is due to environment light, and the distance between the
LEDs and the light sensor.
We concluded that the performance of the application depends on several factors:
(1) The brightness of the LEDs: (2) The sensitivity of the light sensor with respect to
the light received from the LEDs; (3) The surrounding light (in Fig. 7 it can be seen
that the light sensor has a control for graduating the sensitivity in the presence of high
or low light); and (4) The alignment between the LEDs and the light sensor influences
the reception, and consequently the quality of the image and text recovery.
5.1 Future Improvements

A list of improvements to enlarge the efficiency of this application are the following:
(1) Increasing the number and/or power of the LEDs to augment the reception distance;
(2) Using different modulation schemes, and (3) Implement and include a continuous
synchronization mechanism to improve the LED-to-light sensor VLC transmission-
reception link.
Acknowledgments. This work was supported by Instituto Politecnico Nacional Project SIP-IPN
20196133.
References
1. Haas, H.: Wireless data from every light bulb. TED: Ideas worth spreading (2011). http://
www.ted.com/talks/harald_haas_wireless_data_from_every_light_bulb
2. Tsonev, D., Videv, S., Haas, H.: Light fidelity (Li-Fi): towards all-optical networking. In:
Proceedings of SPIE (Broadband Access Communication Technologies VIII), vol. 9007, no.
2 (2013). https://doi.org/10.1117/12.2044649
3. Sherman, J.: How LED light bulbs could replace Wi-Fi. Digital Trends (2016). http://www.
digitaltrends.com/mobile/light-bulb-li-fi-wireless-internet/
4. Haas, H.: High-speed wireless networking using visible light. SPIE Newsroom. https://doi.
org/10.1117/2.1201304.004773, http://www.spie.org/newsroom/4773-high–speed-wireless-
networking-using-visible-light. Accessed March 2016
5. Vincent, J.: Li-Fi revolution: internet connections using light bulbs are 250 times faster than
broadband (2016). http://www.independent.co.uk/news/science/li-fi-revolution-internet-
connections-using-light-bulbs-are-250-times-faster-than-broadband-8909320.html
6. Location Awareness Wikipedia (2018). https://en.wikipedia.org/wiki/Location_awareness
7. Jovicic, A., Junyi, L., Richardson, T.: Visible light communication: opportunities, challenges
and the path to market (2013). https://doi.org/10.1109/mcom.2013.6685754, http://
ieeexplore.ieee.org/document/6685754/
8. Mahendran, R.: Integrated LiFi (Light Fidelity) for smart communication through
illumination. In: International Conference on Advanced Communication Control and
Computing Technologies (ICACCCT), pp. 53–56. https://doi.org/10.1109/icaccct.2016.
7831599, https://ieeexplore.ieee.org/document/7831599/?reload=true
9. Shah, V., et al.: 2D image transmission using light fidelity technology. Int. J. Innov. Adv.
Comput. Sci. (IJIACS) 4(4), 121–126 (2015). https://www.academicscience.co.in/admin/
resources/project/paper/f201504241429887963.pdf, ISSN 2347-8616
10. Lee, K., Park, H.: Modulations for visible light communications with dimming control. IEEE
Photonics Technol. Lett. 23(16) (2011). https://doi.org/10.1109/lpt.2011.2157676, http://
ieeexplore.ieee.org/document/5773477/
11. Perfecto, G.C., Godoy, R.S., Rios, A.A.: Transmision de imagenes de Matlab a Arduino via
Puerto Serial. Boletín UPIITA-IPN, 644 CyT, No. 52 (2016). http://www.boletin.upiita.ipn.
mx/index.php/ciencia/644-cyt-numero-52/1234-transmision-de-imagenes-de-matlab-a-
arduino-via-puerto-serial
12. LittleBits. https://www.littlebits.cc/
13. Pi, R.: Raspberry Pi input and output pin voltage and current capability. Mosaic
Documentation Web. http://www.mosaic-industries.com/embedded-systems/microcontroller-
projects/raspberry-pi/gpio-pin-electrical-specifications
14. Vasuja, M., et al.: Image transmission using Li-Fi. In: IEEE Conference Publications (2018).
https://ieeexplore.ieee.org/document/8473033
15. Raju, K.S., et al.: PC to PC transfer of text, images using visible light communication (VLC).
Int. J. Adv. Eng. Manag. Sci. (IJAEMS) (2017). https://dx.doi.org/10.24001/ijaems.3.5.6
16. Hamand, A., Kuntawar, S.: An approach towards high speed communication using Li-Fi
technology. IOSR J. Electron. Commun. Eng. (2018). http://www.iosrjournals.org/iosr-jece/
papers/Vol.%2013%20Issue%204/Version-1/K1304017275.pdf
17. Chen, M., et al.: Real-time video transmission of visible light communication based on LED.
Int. J. Commun. Netw. Syst. Sci. 10(8B) (2017). https://doi.org/10.4236/ijcns.2017.108b007,
https://www.scirp.org/journal/PaperInformation.aspx?PaperID=78373
18. Costa Mari, J.: Visible light communication for Internet of things. Trabajo final de grado,
Universitat Politecnica de Catalunya (2018). https://upcommons.upc.edu/handle/2117/
117811
12–18 GHz Microwave Frequency Band
Microstrip Patch Antenna Design for the Radio
Implant Medical Devices Application
Salvador Ricardo Meneses González(&)

and Rita Trinidad Rodríguez Márquez
Instituto Politécnico Nacional, Escuela Superior de Ingeniería Mecánica y

Eléctrica, Unidad Zacatenco, 07738 Mexico City, Mexico
rmenesesg@ipn.mx
Abstract. The implantation of radio implant medical devices (IMD) into the
human body are becoming smaller, therefore and because to the high degree of
miniaturization and an efficient wireless link, antennas extremely small, which
satisfies gain, bandwidth, impedance, and low cost are demanded. This work
proposes a wide bandwidth ranging from 12 GHz to 18 GHz Microstrip Patch
Antenna, describing the design, simulation, implementation and measurement,
in order to be used on the IMD’s.
Keywords: Microstrip antenna Microwave band Implant medical device
1 Introduction
Biomedical telemetry permits the measurement of physiological signals at a distance

through wireless communication technologies, that is, implantable medical devices
IMDs are implanted inside the patient’s body and used for diagnostic, monitoring,
and/or therapeutic applications, but the human body is a lossy material for electro-
magnetic waves.
This means the body converts electric fields into heat, that is, the body absorbs
energy from electromagnetic waves and part of this energy is collected by the antenna,
an important element of the IMD, consequently, when an antenna is placed near the
body, the result is a large reduction of the antenna efficiency, this power loss is a
function of frequency which value is in accordance with the permittivity and con-
ductivity of the medium where the electromagnetic wave propagates, in this case,
human skin, fat, muscle, and due to its structure, these ones have different attenuation
factor, that is, human body tissue is a great attenuator [1], which increases with fre-
quency. In this sense, these devices which use small antennas have become in a part of
body-centric wireless communications.
Several different antenna designs have been considered for medical implants, for
instance, [2] evaluates microstrip configurations, spiral and serpentine antennas were
simulated in order to compare their resonant frequencies, concluding that, serpentine
antenna is electrically shorter than the spiral antenna, [3–6] report challenges, limita-
tions and benefits of different kind of antennas. Thus, it is known that the antenna size

https://doi.org/10.1007/978-3-030-33229-7_10
12–18 GHz Microwave Frequency Band Microstrip Patch Antenna Design 111
with respect to the wavelength is the most important parameter influencing the
antenna’s radiation [3], and a relation between antenna size and its radiation charac-
teristics is reported in [7] concluded that geometrically small but efficient antennas are
required for high operating frequencies, and the antenna efficiency significantly
decreases if lower operating frequencies are used. Likewise, implantable antennas
operating at much higher frequencies (31.5 GHz) have also been reported in the lit-
erature [8].
On the other hand, inductive links formerly have been employed for the wireless
telemetry of IMDs [9–11], but this kind of inductive low-frequency links shows low
data rates, short communication range and unwanted emissions. Actually ITU-R
Recommendation SA.1346 (ITU-R 1998) which outlined the use of the 402.0–
405.0 MHz frequency band for medical implant communication services (MICS), in
this sense, at 402 MHz–405 MHz, the wavelength of an electromagnetic wave is
approximately 74 cm, as we know, the size of the microstrip antenna is directly pro-
portional to the wavelength at the operating frequency, consequently an antenna with
dimensions comparable to this wavelength cannot be used for an implant [12], so,
different antenna structures that need to be addressed. In this sense, due to higher
operation frequencies allow use of smaller-sized antennas and components, and without
wanting to enter a conflict with ITU-R regulations, for this purpose the designed
antenna in this work, is a just a proposal to be considered, due to wavelength corre-
sponding to this frequency band (12–18 GHz), ranges between 25 mm to 16 mm.
Only for the sole purpose to observe the dimensions of the proposed microwave
band patch antenna, this one was placed close to a RFID UHF Band meander antenna
(915 MHz) [13], (see Fig. 1).
Fig. 1. (a) RFID UHF Band meander antenna (915 MHz) and (b) Proposed microwave band
patch antenna (12–18 GHz).
Therefore, in view of the importance of their health role and because the body
directly degrades the performance of the antenna, these kind of antennas should be
safety, lightweight, extremely small, robust structure, it must be considered the inter-
action with the human body, in this sense, there are complex factors involved that have
112 S. R. Meneses González and R. T. Rodríguez Márquez
to be considered when dealing with this issue; [14] describes considerations related to
the design and performance of implantable patch antennas. In this work, first part, it is
limited to selection of the frequency band of operation, miniaturization of the antenna,
the other issues relative to biocompatibility of the antenna, will be supported on a
further work (part 2) in this area.
In this way, in this work, we propose a small slotted patch antenna operating in 12–
18 GHz Microwave Band, in order to reduce the size antenna, increase the bandwidth,
and improve the quality communication of the implantable antenna with exterior
monitor/control equipment.
To determine the performance of design parameter, as impedance, resonance fre-
quency, radiation pattern, wideband, etc., CST software (Computer Simulation Tech-
nology) has been used, in the same way, experimental tests have been applied.
The paper is organized as follows: Sect. 2 describes a brief tag antenna design
basis, discusses simulation, Sect. 3 describes measurements results, and finalizing the
present work with conclusions and references.
2 Antenna Design
2.1 Design Basis

The patch antenna is generally made of conducting material such as cooper and the
substrate material should have dielectric constant, and it bases her behavior on the way
in which the electromagnetic waves propagates through the antenna structure. Theses
ones are reflected by the ground plane, traveling on the dielectric surface, guided by the
conductive strip, when they reach the point between the face substrate an the free space,
part of the electromagnetic energy is reflected, and part of it, it is radiated to free space;
so, in order to increase the antenna efficiency, the radiated energy should be higher than
the reflected energy.
In this work, we propose a small patch antenna which consists of a rectangular
patch fed by a microstrip transmission line, operating at 12–18 GHz frequency band, in
order to miniaturization of the antenna’s occupied volume/size and achieve high effi-
ciency and large bandwidth. Let us remember that thick substrates with low permit-
tivity result in antenna designs with high efficiency and large bandwidths, thin
substrates with high permittivity lead to small antennas but with a lower bandwidth and
high radiation loss [15]; during the simulation stage, we have made a tradeoff between
substrate thickness and permittivity for the purpose to define the kind of material,
according to within our means in a cost-effective way, for that, we have used and built
the prototype antenna on a duroid substrate, which electric permittivity is er = 2.2 and
loss tangent d = 0.0009.
On the other hand, patch antennas, not all of them preserving the characteristics of
the device in a frequency band broad enough and some of them are complicated in
terms of design and have the disadvantage that bandwidth is limited, so, in order to
enhance the antenna, some modifications on the structure are applied, based on slots to
redistribute the current density. In fact, the antenna has been designed to be more
simple, low cost, wide bandwidth, without the need of altering of substantial changes to
the structure, maintain the small size.
Considering a design frequency equal to 15 GHz, this being the core of the pro-
posed band, the wavelength is equal to k = 20 mm. As we know, the size of the
microstrip antenna is directly proportional to the wavelength at the operating fre-
quency, in this sense, the following expressions [16] have been used, in such a way
that, the substrate thickness h = 1.524 mm, strip width, W = 9 mm, which is
approximately equal to k=2.
Structure geometry and dimensions are shown in Figure.
Strip width:
rffiffiffiffiffiffiffiffiffiffiffiffi
1 2
W ¼ pffiffiffiffiffiffiffiffiffi ð1Þ
2fr 0 l0 r þ 1
Strip length:
1
L¼ pffiffi pffiffiffiffiffiffiffiffiffi 2DL ð2Þ
2fr e e0 l0
Effective Permittivity:
0 1
r þ 1 r 1 B 1 C
e ¼ þ @qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA ð3Þ
2 2 1 þ 12 h
W
where:
fr , resonance frequency
er , relative permittivity
h, substrate thickness.
2.2 Antenna Simulation

Important parameters that define the antenna performance are the reflection coefficient
or return loss, which is called Parameter, S11, that, describes the performance antenna,
for instance, if S11 = −10 dB, this implies 90% of power is delivered to the antenna,
and 10% is the reflected power, when this value crosses the power magnitude line
−10 dB value a certain frequency, this is the resonance frequency or bandwidth on a
frequency range, if S11 = 0 dB, then all the power is reflected from the antenna and
nothing is radiated. The other parameter is the radiation pattern (Fig. 2).
In order to achieve these parameters, the computer Simulation Technology Soft-
ware [17] has been used to simulate the designed antenna (see Fig. 3), resulting the
Magnitude vs. Frequency simulation graphic, parameter S11 (see Fig. 4).
It is possible to observe that from the magnitude vs. frequency simulation graphic,
resonance frequency value is approximately equal to 15 GHz, bandwidth approxi-
mately equal to 1.5 GHz, from 13 GHz to 14.5 GHz and from the simulation trials
Fig. 2. Patch antenna dimensions.
Fig. 3. Antenna structure simulation.
results gives the chance to identify the necessary dimension modifications, in order to
adjust the resonance frequency value along the 12–18 GHz. Simulated radiation pattern
(see Fig. 5) shows the vertical plane (E| vs. h), called E-Plane and the 3D simulation
pattern (See Fig. 6).
In the same way, it is possible to observe an omnidirectional performance, with no
nulls, concentrated electromagnetic power on the −90°–90° plane, that is, the front side
of the antenna, due to ground plane to its rear. This radiation pattern geometry is an
Fig. 4. Magnitude vs. Frequency (simulation).
Fig. 5. Radiation pattern graphic, Plane-E (simulation).

Fig. 6. 3D radiation pattern graphic (simulation).
advantage, because of point to point communication is required for this application and
certain area close to the body skin can be covered.
3 Experimentation
In order to build the prototype, we have used Copper slotted by tracks engraved in
DUROID-type substrate with a width of 1.5 mm, which electric permittivity is er =
2.2, and a SMA connector. It should be noted that we have used and ordinary SMA
connector, due to our means in a cost-effective way. Use a better-quality and appro-
priate connector to work at this frequency band, improve the antenna efficiency (see
Fig. 7).
In the same way, we have used a Vector Network Analyzer ZVB 40 is calibrated in
the band 10 GHz–20 GHz, at short circuit, opened circuit and matching network (see
Fig. 8) in order to measure S11 parameter.
From the magnitude vs. frequency graphic achieved (see Fig. 9), it is possible to
observe that the maximum antenna efficiency is achieved at 12.86 GHz, resonance
frequency, point M2 and at approximately 13.2 GHz is the minimum efficiency. The
bandwidth is approximately equal to 6 GHz, starting in 12.11 GHz (point M1) and
ending in 18.5 GHz, that is, the designed antenna is a wideband antenna, bandwidth
enough to support high data rate.
On the other hand, literature review present other architectures antennas, in the
MICS (402.0–405.0 MHz) and ISM (433.1–434.8, 868.0–868.6 and 902.8–
928.0 MHz) bands [18]. The proposed antenna is novel, useful, innovative antenna, in
the microwave band (12–18 GHz), small dimensions and wide bandwidth.
Fig. 7. Prototype antenna.
Fig. 8. Prototype antenna under test.
3.1 Limitation
The method used to measure and calculate the gain antenna and radiation pattern,
consists of install the prototype antenna, into an anechoic chamber, acting as receiver
antenna, and a known second antenna, a commercial antenna acting as a transmitter
Fig. 9. Magnitude vs. Frequency, S11 parameter (prototype antenna under experimental test).
antenna, both of them installed into the anechoic chamber, spaced 2.8 m apart, keeping
a line of sight (see Fig. 10), however, actually, we are not be able to carry out this
measurement due to our laboratory does not have capabilities to experiment at this band
frequency (transmitter and receiver equipment).
Fig. 10. Gain antenna measurement method.

4 Conclusion
A microstrip patch antenna array has been designed, meeting the resonance frequency,
small size (millimetric), exhibiting a wide bandwidth (12–18 GHz), an appropriate
geometry of the radiation pattern, low cost, and easily construction.
There are differences between simulation and measurement results on the S11,
respect to the maximum resonance frequency, 15 GHz and 12.86 GHz, respectively,
due to certain construction defects, for instance, bad soldier, SMA connector, or adjust
the exact dimensions, however, the experimental bandwidth result was far better than
the simulated bandwidth result, due to the simulation bandwidth result is equal to
1.5 GHz and the experimental bandwidth result is approximately equal to 6 GHz.
The second part of this work consists of to continue the technical stage (gain
measurement), as soon as we get the chance, as soon as the appropriate equipment at
this band is in place, and, verify the antenna effectiveness (distance, penetration depth),
testing the antenna performance on/under human or animal skin, it will be necessary
the support biomedical research.
Finally, due the obtained results, the proposed antenna is a novel, useful and
innovative antenna for IMD’s application, in the microwave band, wide bandwidth,
small dimensions, easily construction, meeting the resonance frequencies, input
impedance requirements, an appropriate geometry of the radiation pattern, so that, we
recommend this kind of small antenna as well as the proposed frequency band to be
applied to Radio Implant Medical Devices corresponding to the health sector.
References
1. Xuyang, L.: Body Matched Antennas for Microwave Medical Applications. KIT Scientific
Publishing, Karlsruhe (2014)
2. Soontornpipit, P., Furse, C.M., Chung, Y.C.: Design of implantable microstrip antenna for
communication with medical implants. IEEE Trans. Microw. Theory Tech. 52(8), 1944–
1951 (2004). Part 2
3. Kim, J., Rahmat-Samii, Y.: Implanted antennas inside a human body: simulations, designs,
and characterizations. IEEE Trans. Microw. Theory Tech. 52(8), 1934–1943 (2004). Part 2
4. Norris, M., Richard, J.-D.: Subminiature antenna design for wireless implants. In:
Proceedings of the IET Seminar on Antennas and Propagation for Body-Centric Wireless
Communications, London, UK, April 2007, pp. 57–62 (2007)
5. Hall, P.S., Hao, Y.: Antennas and Propagation for Body-centric Wireless Communications.
Artech House, Norwood (2006)
6. Skrivervik, A.K., Zürcher, J.F., Staub, O., Mosig, J.R.: PCS antenna design: the challenge of
miniaturization. IEEE Antenna Propag. Mag. 43(4), 12–27 (2001)
7. Sánchez-Fernández, C.J., Quavado-Teruel, O., Requena-Carrión, J., Inclán-Sánchez, L.,
Rajo-Iglesias, E.: Dual-band microstrip patch antenna based on short 0 circuited ring and
spiral resonators for implantable medical devices. IET Microw. Antennas Propag. 4, 1048–
1055 (2010)
8. Ahmed, Y., Hao, Y., Parini, C.: A 31.5 GHz patch antenna design for medical implants. Int.
J. Antennas Propag. 2008, 6, Article ID 167980 (2008). Hindawi Publishing Corporation
9. Freedonia: Implantable Medical Devices – Industry study with forecast for 2015 & 2020.
The Freedonia Group, Study #2852 (2012)
10. Valdastri, P., Menciassi, A., Arena, A., Caccamo, C., Dario, P.: An implantable telemetry
platform system for in vivo monitoring of physiological parameters. IEEE Trans. Inform.
Technol. Biomed. 8, 271 (2004)
11. Baghel, S., Yadav, N.K.: A review on micro electronic pill. Int. J. Res. Appl. Sci. Eng.
Technol. (IJRASET), 3(VI) (2015)
12. Kiourti, A., Nikita, K.S.: Implanted antennas in biomedical telemetry. In: Chen, Z.N., Liu,
D., Nakano, H., Qing, X., Zwick, T. (eds.) Handbook of Antenna Technologies, pp. 2613–
2652. Springer, Singapore (2016). https://doi.org/10.1007/978-981-4560-44-3_94
13. Meneses, R., Montes, L., Morales, M.: A small note about RFID technology applied to
public safety. Part 1: RFID tag antenna design. In: Conference: Power, Electronics and
Computing (ROPEC), 2013 IEEE International Autumn Meeting 2013, Ixtapa, Gro. México,
November 2013
14. Li, X.: Body Matched Antennas for Microwave Medical Applications. Forschungsberichte
aus dem Institut für Höchstfrequenztechnik und Elektronik (IHE) der Universität Karlsruhe
(TH). KIT Scientific Publishing (2015). ISSN 0942-2935
15. Madhav, B.T.P., et al.: Multiband slot aperture stacked patch antenna for wireless
communication applications. Int. J. Comput. Aided Eng. Technol. 8(4), 413–423 (2016)
16. Wong, K.: Compact and Broadband Microstrip Antennas. Wiley, Hoboken (2002)
17. CST Homepage. https://www.cst.com/Academia/Student-Edition
18. Kiourti, A., Nikita, K.S.: Performance of miniature implantable antennas for medical
telemetry at 402, 433, 868 and 915 MHz. In: Godara, B., Nikita, K.S. (eds.) MobiHealth
2012. LNICST, vol. 61, pp. 122–129. Springer, Heidelberg (2013). https://doi.org/10.1007/
978-3-642-37893-5_14
Analytics for Basic Products in Mexico
Paul Millán(&) and Félix Mata
Instituto Politécnico Nacional, Unidad Profesional Interdisciplinaria en

Ingeniería y Tecnologías Avanzadas, Avenida Instituto Politécnico Nacional
No. 2580, Col Barrio la Laguna Ticomán, Gustavo A. Madero,
07340 Mexico City, Mexico
luapmg@outlook.es, mmatar@ipn.mx
Abstract. In this research, it was designed a methodology to forecast prices of

products In the Mexican Republic. The dataset is composed of basic basket
products. The work consists of analyzing open and mixed data of this dataset.
The approach is centered on studying how is the behavior in time and location
domains for three products, tuna, detergent, and milk. The data ranges for five
years. Neural networks were used to analyze data, and several experiments of
price forecast were issued using different granularity levels. The regression
models were validated using two traditional approaches of the machine learning
area, coefficient of determination, and mean absolute error. The experiments
showed that the price of basic products varies by zone and it is possible to give a
forecast with a percentage of 80% of precision.
Keywords: Forecasting Data analytics Data science Big Data Machine

learning
1 Introduction
Presently, the data of basket products have been studied from a wide range of per-
spectives mainly in financial and market aspects. Traditionally these approaches are
dominated for statistical approaches. However, new opportunities of analyzing and
studying these phenomenons from the perspective of data science, Big Data, and
machine learning ara available today to discover or identify insights in such data.
For another hand, according to the National Survey of Household Income and
Expenses (INEGI by its acronym in Spanish) [1]. The average household expenditure
in Mexico is invested in products of a basic basket. In light of this knowledge, it is
natural to be interested in analyzing the behavior of these products. The possible results
of the data analysis of these products have a valuable and useful character. Future price
forecasts can help commercial companies in the elaboration of control and financial
planning processes. Governments can also benefit from this type of study for the
creation of public policies.

https://doi.org/10.1007/978-3-030-33229-7_11
122 P. Millán and F. Mata
2 Methodology
The data methodology proposed is composed of 4 phases: (1) ETL process, (2) Data
Continuity, (3) Categorical Data transformation and (4) Feature Selection. Each of one
are developed in the following sections.
2.1 ETL Process

The first phase of the methodology is based on the process of extraction, transfor-
mation, and loading (ETL).
The data used in this research are open and provided for a Mexican organism
denominated PROFECO [2]. The data is available in CSV and JSON format. The data
period includes from 2011 to 2016, in the Mexican Republic, with more than 62
million records. The nature of data is mixed, numerical, and categorical. The data was
cleaned, eliminating records containing empty, null, or ‘N/A’ values. The data of the
address (location) variable were homogenized since the oldest records only contained
the street and the number, while the most current ones also contained the ‘barrio’. An
identity location procedure is running in order to match all the address. Two original
variables, called latitude and longitude, were eliminated since it was not planned to
work at this level of geographic granularity. Three new integer variables called day,
month, and year were created, based on the original date variable. Finally, the Date-
Time type variable is transformed into a continuous floating type value to work in the
next process with them.
2.2 Data Continuity

It is necessary to guarantee certain continuity in the records so that there is homoge-
nized information, which will help to create and evaluate better regression models as
well as compare products taking into account the variability of their price over time.
For the purpose of this research, a combination will be when a number of records have
the same values taking into account a certain number of variables, except for this rule
the date and price variables. In this way when there are two records that are the same
except for the date and the price, it will be said that there are two records of this
combination.
In the data of this research there was a serious problem of continuity over time, as
an example, see the Fig. 1, where this problem can be seen for a combination of three
variables of the milk powder product.
It required to create a new training dataset, where those combinations that met a
certain condition of continuity were taken, in this case, it was proposed to keep those
combinations that appeared at least 5 years and at least 8 months for each year. In the
example in the previous Fig. 1, only the first combination would be selected, and the
others would be omitted.
Analytics for Basic Products in Mexico 123
Fig. 1. Time continuity of milk powder records in the dimensions of product, presentation,
brand.
2.3 Categorical Data Transformation

Currently, most data mining and machine learning techniques need numerical data.
Therefore, categorical data such as those present in this investigation should be
transformed into numerical values to which it is possible to apply these techniques. In
this methodology it is recommended to use a binarization technique of type one against
all, for example, One Hot Encoding, see Table 1. This technique is based on converting
every possible value of a categorical attribute into a new column. The disadvantage is
that this technique increases the number of columns in the dataset.
Table 1. Binarization sample: one versus all.

Brand Price (MXN) Brand Price (MXN)
Dolores Herdez Tuny
Dolores 11.5 1 0 0 11.5
Herdez 11.9 0 1 0 11.9
Tuny 12 0 0 1 12
2.4 Feature Selection

One of the problems that were constantly presented during this investigation was the
high computational cost, which translated into long periods to process the data, as well as
for the stages of machine learning, mainly in the training stage. This was due in large part
to the increase in dimensions that occurs when transforming categorical to numerical
data through binarization. It was necessary to use feature selection algorithms to select
the most important and discard those that provided little information when trying to
predict the price. It was necessary to use feature selection algorithms to select the most
important and discard those that provided little information when trying to predict the
price. This was achieved by using cloud computing using the mutual information
algorithm which can work with categorical data. Analyzing the characteristics using all
the basic products we have as a result the Table 2 showing the most important variables.
Table 2. Feature selection for all basic products.

Presentation Product Brand Category Address Tradename
1.5460 1.1770 1.0930 0.5070 0.0470 0.0390
The above is consistent since the products are being taken in a grouped manner,
therefore characteristics such as product and presentation are very important. When the
same feature selection algorithm is applied to certain products and certain presentations
individually, other features such as address become more important. It is shown in
Table 3.
Table 3. Feature selection for certain products.

Product Presentation Brand Address Tradename Year Commercial
Chain
Milk 0 1.0935 0.4702 0.2432 0.1401 0.1811 0.0748
powder
Tuna 0 0.1216 0.0931 0.3733 0.2770 0.0965 0.2055
Detergent 0.292 1.1890 1.0550 0.152 0.1270 0.0090 0.09900
2.5 Parameter Selection and Model Generation

The regression algorithm used in this investigation is MLPRegressor [3–6]. To select
parameters, the exhaustive search algorithm GridsearchCV was used, in addition to the
latter, TimeSeriesSplit was used as a cross-validation technique with a number of
divisions defined to 5. The previous algorithms belong to the library Scikit-learn [7] in
python. The MLPRegressor class implements a multilayer perceptron (MLP) that trains
using backward propagation [8, 9] with no activation function in the output layer,
which can also be seen as the function of identity as a function of activation. Use the
square error as the loss function, and the output is a set of continuous values.
MLPRegressor uses the alpha parameter for the term regularization (L2 regularization)
that helps avoid over-setting [10]. It makes use of a stochastic gradient-based optimizer
called adam [11].
The MinMaxScaler algorithm is used to scale all values to a range between 0 and 1.
The selection of the alpha parameter is very important since this parameter helps to
control the over-adjustment, it is recommended to use a range of [1.e−01, 1.e−02, 1.e
−03, 1.e−04, 1.e−05, 1.e−06] for the MLPRegressor algorithm [10]. As for the size of
the neural network, it was tested with different sizes using GridsearchCV. The first size
equal to the number of real entries, the second equal to the number of entries after
binarizing the categorical data. The above was tested for each case from a single hidden
layer to three.
2.6 Measure Performance with Grouped Products

The coefficient of determination with which model´s learning is being measured uses
the variance of the set of samples to be defined, see formula 1. Now, the variance is the
square of the standard deviation, therefore it is necessary to pay special attention to the
cause of the latter.
Pm
ðyi ^yi Þ2
R2 ¼ Pi¼1
m
ð1Þ
i¼1 ðyi yi Þ 2
Looking at the previous formula it can be seen that the coefficient of determination
is based on knowing how good the predictions generated by the model are, compared
with simply having used the average of the samples as a prediction. If there is a larger
standard deviation, there will be a larger error in the lower part of the equation, which
will minimize the error produced by the model. Therefore, it is vitally important to
know if this standard deviation is due to the variation of a product over time or if it is
due to grouping, for example, different products and presentations. To clearly visualize
this situation, a detergent model created from a dataset whit different presentations are
used as an example. The presentations are as follows: bottle 1 L liquid, bottle 750 ml
liquid, bag 500 g powder, bag 900 g powder, bag 1 kg powder, bag 5 kg powder. The
presentation in bag 5 kg powder stands out in size from others, in the same way, its
price is higher. In the test stage, the model demonstrated know to differentiate between
this presentation and the others, clearly observing a distance between the forecasts,
always because of the presentation of bag 5 kg powder that generated higher prices.
The above can be seen in the Fig. 2.
Fig. 2. Error comparison with the detergent product.
The predictions shown in Fig. 2 come from a model that obtained a coefficient of
determination of 0.98.
2.7 Measuring the Model’s Performance with Ungrouped Products

The models generated with datasets of ungrouped products will obtain a coefficient of
determination that will reflect exclusively how good the predictions of a particular
product and presentation are, taking into account their own variance over time.
Fig. 3. Error comparison with the model of ultra-pasteurized milk, partially skimmed, box 1 L
Lala light milk.
The model presented in Fig. 3 obtained a coefficient of determination of 0.7, this

value is less than that obtained in the previous section which was 0.98. However, it can
be seen in Fig. 4 that the forecasts are much closer to real prices, many of errors are in
the order of cents.
Fig. 4. Test for ultra-pasteurized milk, partially skimmed, box 1 L brand Lala Light in
Chedraui, Toluca, Mexico, Alfredo del Mazo 705, col. Tlacopa CP 50010
3 Results
A model and predictions were created of ultra-pasteurized milk, partially skimmed, box
1 L brand Lala Light. A brief description of the dataset is presented in Table 4.
Table 4. Description ultra-pasteurized milk, partially skimmed, box 1 L brand Lala Light.
Number of records 39939
Mean 14.2270
Standard deviation 0.9732
Minimum 8.0000
25% 13.5000
50% 14.3000
75% 14.9000
Maximum 25.9000
The characteristics selected in this model are the following: date, year, commercial
name, address, municipality, state, commercial chain and month. The result of the
parameter selection can be seen in the Table 5.
Table 5. Results of the selection of parameters for ultra-pasteurized milk, partially skimmed,
box 1 L brand Lala Light.
Alpha Network size Ranking
0.001 (527, 527, 527) 1
0.1 (8) 2
0.01 (527, 527) 3
0.000001 (527, 527, 527) 4
The training period was from January-2011 to June-2015. The yield obtained was
0.1722 of the coefficient of determination and an average absolute error of 0.5131, both
in the test stage, which covered from June 2015 to April 2016.
Now, an example of the predictions generated during the test stage for an address,
that is, an establishment, in this case of the Chedraui commercial chain in the state of
Toluca. See Fig. 4.
Using this methodology, it was discovered that the prices of basic products in this
database vary by address, prices are not kept by commercial chains. Even establish-
ments of the same commercial chain in the same state and municipality, present their
own prices, an example can be seen in Table 6. In this case, there is a price difference
of more than 15%.
Table 6. The same detergent product offered at Walmart in two establishments at different
prices on the same date.
Product Presentation Brand Price Date Commercial Address State Municipality
chain
Dish Bag 900 g Salvo 16.91 2011- Walmart Calzada de Distrito Gustavo A.
detergent powder 04-08 Guadalupe 431 Federal Madero
(green)
Dish Bag 900 g Salvo 19.50 2011- Walmart Othon de Distrito Gustavo A.
detergent powder 04-08 Mendizabal Ote Federal Madero
(green) 343
Fig. 5. Location and distance between two Walmart stores.
It is interesting how this variation occurs even in nearby locations, like in this case,
two establishments belonging to the Walmart commercial chain at approximately 3 km
away. It is depicted in Fig. 5.
4 Conclusions and Future Work
We conclude t is possible not only to know the behavior of the products but also to
generate models of prediction of the prices of these products in a particular place and
some period.
It is possible by combining the techniques of exploration and machine learning on
time and location domains. It is applied to categorical attributes such as product,
presentation, brand, and commercial chain, address, among others.
Nevertheless, It is necessary to work on other approaches that allow the transfor-
mation of qualitative or categorical data to numerical data. Considering the conditions:
do not insert bias and not increase the number of columns.
The use of Feature selections algorithm allowed identifying datasets with better
performance than the original ones to produce price predictions. However, it is pro-
posed as future work to add new variables that can later be analyzed with other
algorithms, to discover possible external variables to this dataset that influence the
variation of prices products in Mexico.
References
1. INEGI Homepage. Encuesta Nacional de Ingresos y Gastos de los Hogares (ENIGH) (2014).
https://www.inegi.org.mx/programas/enigh/tradicional/2014/default.html. Accessed May
2019
2. PROFECO Homepage. Quién es Quién en los precios. https://datos.gob.mx/busca/dataset/
quien-es-quien-en-los-precios. Accessed January 2018
3. Hinton, G.E.: Connectionist learning procedures. In: Artificial Intelligence, pp. 185–234
(1990)
4. Glorot, X., Bengio, Y.: Understanding the difficulty of training deep feedforward neural
networks. In: Thirteenth International Conference on Artificial Intelligence and Statistics,
Sardinia, Italy (2010)
5. He, K., Zhang, X., Ren, S., Sun, J.: Delving deep into rectifiers: surpassing human-level
performance on ImageNet classification. In: Proceedings of the IEEE International
Conference on Computer Vision, pp. 1026–1034 (2015)
6. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization (2014)
7. Pedregosa, F., Varoquaux, G., Gramfort, A., et al.: Scikit-learn: machine learning in Python.
J. Mach. Learn. Res. 12, 2825–2830 (2011)
8. Rumelhart, D.E., Hinton, G.E., Williams, R.J.: Learning representations by back-
propagating errors. Cogn. Model. 5(3), 1 (1988)
9. LeCun, Y.A., Bottou, L., Orr, G.B., Müller, K.-R.: Efficient BackProp. In: Montavon, G.,
Orr, G.B., Müller, K.-R. (eds.) Neural Networks: Tricks of the Trade. LNCS, vol. 7700,
pp. 9–48. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-35289-8_3
10. Pedregosa, F.: «Scikit Learn» [En línea]. https://scikit-learn.org/stable/modules/neural\_
networks\_supervised.html\#regression. Accessed July 2019
11. Kingma, B.J.: Adam: a method for stochastic optimization. In: 3rd International Conference
for Learning Representations, San Diego (2014)
Design and Development of Photovoltaic
Power Meter
Jaime Vega Pérez1(&), Blanca García2, and Nayeli Vega García3

1
Escuela Superior de Ingeniería Mecánica y Eléctrica,
Instituto Politécnico Nacional (IPN), San José Ticoman,
jvegape@ipn.mx
2
Centro de Estudios Científicos y Tecnológicos No. 1 Gonzalo Vázquez Vela,
Instituto Politécnico Nacional (IPN), Pueblo de San Juan de Aragón,
blgarcia16@yahoo.com
3
Escuela Superior de Cómputo, Instituto Politécnico Nacional (IPN),
Zacatenco, 07738 Mexico City, Mexico
nvegag0126@gmail.com
Abstract. The design, construction and test of an electronic system in order to

measure of photovoltaic (PV) module power are reported. An interface elec-
tronic system was developed to couple the PV module toward an electronic
control target with a personal computer. Samplings of the PV module current
and voltage are captured analogically. The electronic system amplifies the
electrical signal and converts to digital signal and also software was designed
based in Matlab. After using this software the digital current and voltage signals
are processed by the laptop to determinate the PV module power. The output
short circuit current, open circuit voltage and maximum power of PV module
were measured. In addition, the output current-voltage and power-voltage curves
are achieved. This measurement system has advantages of simple, cost effective
and applicable in the place where the modules are installed to measure the PV
modules. Experimental results were done for different levels of incident solar
power to measure capacity of PV modules installed in CINVESTAV-IPN,
Mexico. The prototype developed has tolerance of 1.5% but it can be improved
using precision electronic devices. We have obtained the I-V curve, P-V curve
and maximum power point of the PV modules with the help of this prototype.
Keywords: Photovoltaic module Photovoltaic power meter Maximum

power
1 Introduction
Till now, the greatest amount of electricity used by man is extracted from natural fossils
fuels. The energy generation by fossils fuels has concerned due to its pollution on eco-
system and limited amount available. So that it is now necessary to look for alternating
energy sources that can supply energy without producing harmful gases for long
period. Among different alternative sources of energy, solar energy is sustainable and

https://doi.org/10.1007/978-3-030-33229-7_12
Design and Development of Photovoltaic Power Meter 131
renewable energy, which can be converted directly from sun energy to electrical energy
by photovoltaic (PV) cells and modules [1]. The photovoltaic (PV) cell is an electronic
device, which transforms the visible light into direct current electrical energy. The
output power depends on area of cells, incident light level, efficiency but it is reduced
for temperature increase. Due to the photovoltaic cell is a low voltage device (0.56 V)
then photovoltaic modules are built that are integrated by several electrically connected
cells in series to increase the voltage.
The basic module is designed to couple with an automotive type battery, its output
voltage in the open circuit of the order of 17 to 19 V so that it can supply power to the
battery. Its output power is variable from 20 W and depends of cell size.
The manufacturers are fabricating PV cells and modules of mono crystalline and
polycrystalline silicon principally although many other semiconductor materials can be
used [1]. The basic PV module is integrated by 36 PV cells connected in series con-
nected to automotive battery of 12 V, but with least time PV modules with higher output
power and voltages are being manufactured [2]. The PV cell or module is electronic
device that converts the light energy from solar energy into the electrical energy or direct
current (DC). The PV cell is a semiconductor diode the more common is done with
silicon and phosphorus although there are many semiconductors materials to make PV
cells [2]. The output current of photovoltaic module is represented by Eq. 1.
h qðVc þ Ic Rs Þ i I R þV
c s c
Ic ¼ I0 e KT 1 þ þ Ig ð1Þ
Rf
The PV module output is represented mathematically by the equation of rectifier

diode but guessing the serial resistance and shunt resistance as it is showed in Eq. 2.
qðVo þ Io Rs Þ I R þV
o s o
Po ¼ Vo Ig I0 e KT 1 ð2Þ
Rf
From Eq. 2, Po is the output power, Vo is the output voltage, Ig is the generated
current, q is the electron charge, Io is the output current, K is the Boltzmann constant,
T is the temperature, Rs is the series resistance [3]. When both are solved changing the
voltage Vc from open circuit condition to short circuit condition, all the values of power
and voltage are obtained, in order to get the current-voltage (I-V) and power-voltage
(P-V) curves of the PV module (see Fig. 1).
The characteristic I-V and PV curves (see Fig. 1) of the photovoltaic module are
very important because they have the information of their electrical variables. The
maximum output power is used when a photovoltaic system is designed. But the power
of the module is variable because it depends mainly on the environmental conditions of
irradiance level, wind speed and module temperature. The modules are electrically
characterized by the manufacturer in standard conditions (i.e. solar power of
1000 W/m2, temperature of 25 °C) where in real working conditions the photovoltaic
module is subjected to environmental changes all time, then it is necessary the PV
module output maximum power must be measured on installation place.
According to the graph in Fig. 1, the I-V curve of solar module is nonlinear, and
then it is not easy to locate the maximum power point (MPP). So, it is necessary to
132 J. Vega Pérez et al.
Fig. 1. Graphic representation of the Eqs. 1 and 2
develop techniques and prototypes to measure the maximum power. Several

researchers have reported different methods to obtain the characteristic curves of the
photovoltaic module to track the maximum power (MPPT).
Ashish et al. [4] developed an algorithm where proposed a variable step-size P&O
method in which the perturbation step is obtained by a fixed ratio multiplies by the
derivative of the power with respect to the voltage. This P&O method sampled the
voltage and current, and the power can be obtained by multiplying the two values.
When both, power variation and voltage variation, are greater than zero, the OP is on
the left-hand side of the P-V curve and moves toward the MPP. If the power variation is
greater than zero and voltage variation is smaller than zero, then the OP is on the right-
hand side of the P-V curve and moves toward the MPP. Therefore, the direction of
perturbation does not change under these two conditions. This method shortens the
tracking time and reduces the steady-state oscillation as well.
Again, Khatib et al. [5] developed a new I-V curve prediction method using arti-
ficial neural networks, they include records of solar radiation, ambient temperature,
current and voltage characteristics (open circuit voltage and short circuit current) for
different photovoltaic modules by using MatLab software.
Danandeh et al. [6] also explain that there are still two important barriers in the path
of PV cells, relatively expensive cost and low efficiency. Therefore it is completely
necessary to extract maximum power of cells because of economic reasons and
growing demand for energy. The solar cell output voltage and current increased with
increasing the input solar radiation to the solar cell, but the PV cell temperature
increases with increase of the solar radiation and temperature results in a decrease of
both power and efficiency Solimana et al. [7]. Heat sink cooling system was used on the
performance of solar cells because the temperature of the solar cell decreases the solar
cell performance.
Through sampling and calculation, the maximum Power Point is tracked success-
fully by determining the step size according to the present and previous values of
voltage and power variation to strike a balance between tracking time and tracking
accuracy. Peng [8]. To solve the problem, variable-step size MPPT algorithms have
been reported and they concluded the transient response of PI type P&O is the fastest
among all. However, the complexity of this method is also the highest.
Santos et al. [9] proposed a traditional Perturbation and Observation to track the
maximum power point (P&O) MPPT technique with trade-off between step-size,
tracking-time and tracking-accuracy. The traditional P&O method sampled the voltage
and current, and the power can be obtained by multiplying the two values. Farhat et al.
[10]. Developed un algorithm based on the architecture Multi-layer Perceptron
(MLP) to track real time, the maximum power point (MPPT). Maximum Power Point is
a function of irradiance and temperature. In contrast, the PV systems must be designed
to operate at their peak output power to extract and maintain the peak power of the PV
panel even when climatic conditions are unfavorable. The obtained results show that
the use of artificial neural networks present a precise estimate of the maximum power
of PPV compared to the measured values.
Ramana et al. [11] described an MPPT algorithm to track the global peak out of
many local peaks of PV module. The output of the PV array mainly depends on the
irradiance and temperature where the non-uniform irradiance conditions occurs on PV
module due to passing clouds, bird dropping, shadows etc. The MPPT algorithm tracks
the global pick out of the many local peaks. The input to the PV array is irradiance and
temperature of four modules and the output is the duty ratio. Pulse Width Modulator
receives the duty ratio as input and generates pulses to the MOSFET of boost converter.
Battery is used as load.
Wang et al. [12] defined an improved particle swarm optimization algorithm is
proposed to track the maximum power point of the PV array when the PV array is
partially shaded by object occlusion. The model of PV array is established, and sim-
ulation is carried out to verify the effectiveness of the algorithm. The method can avoid
the local maximum and track the global maximum power point. Badis et al. [13]
reported the conventional methods such as Perturb and Observe (P&O) and the
Incremental Conductance and fails to extract the global MPP of the PV panel if the PV
generator is partially shaded. They propose an Evolutionary Algorithms (AE), trough
the Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) is better. They
developed an algorithm based in neural networks using MatLab software. Moreover, it
is proven that PSO process mitigates more efficiently power losses caused by PS with
lesser fluctuation and higher steady final output power.
2 Proposal
We propose an integrated prototype by an electronic circuit with two sensors for

current and voltage (see Fig. 2). After they were used, two amplifier steps with inte-
grated circuit LM308. The output signals of this amplifier are sent to analogical-digital
converter. An electronic power circuit was designed [14] to drain up to 10 A which is

coupled to the photovoltaic module, with the combination of analogic integrated cir-
cuits like TL081 and power transistor as MJ5529. A circuit generator was designed
[15] to generate the random time to do the measurement. Tl081 integrated circuit,
transistors as BC547, an electrolytic capacitor and several resistances were used. An
electronic ramp voltage generator circuit was also designed [16] to activate the power
circuit coupled to the PV module, using analogic integrated circuit MC308 combined
with capacitors and resistance to get a time reference in which the electronic system can
do the measurement This circuit was calibrated to get the range of 1 to 5 s so that the
laptop can capture, the storage and processing of the electrical signals found during the
measurement. Figure 3 showed the block diagram of this electronic circuit. Also the
Arduino control target was used to link the electronics circuit with the laptop. For this,
it is necessary to design the appropriate software based in Matlab.
Fig. 2. Block diagram of proposed electronic system
3 Design of Electronic Circuit
The prototype proposed is integrated by electronic circuit, two sensors for current and
voltage. After that, they were used as two amplifier steps with integrated circuit LM308
The output signals of this amplifier are sent to analogical-digital converter. An elec-
tronic power circuit was designed [14] to drain up to 10A which is coupled to the
photovoltaic module, with the combination of analogic integrated circuits like TL081
Fig. 3. Block diagram of the electronic circuit proposed

and power transistor as MJ5529. A circuit generator was designed [15] to generate the
random time to do the measurement. Tl081 integrated circuit, transistors as BC547, an
electrolytic capacitor and several resistances were used. An electronic ramp voltage
generator circuit was also designed [16] to activate the power circuit coupled to the PV
module, using analogic integrated circuit MC308 combined with capacitors and
resistance to get a time reference in which the electronic system can do the measure-
ment This circuit was calibrated to get the range of 1 to 5 s so that the laptop can
capture, the storage and processing of the electrical signals found during the mea-
surement. In Fig. 3 is showed the block diagram of this electronic circuit. Also, the
Arduino control target was used to link the electronics circuit with the laptop. For this,
it is necessary to design the appropriate software based in Matlab.
4 Software Design
For accessing the voltage and current signals, the Arduino electronic card was used
between the developed electronic circuit and the PC and the software was designed
guessing the following logical of the diagram on Fig. 4. The voltage signals were
Fig. 4. Flowchart of proposed method to obtain the PV module maximum power

processed using Matlab software and an Arduino hardware was use in other to capture
the electrical signal of current and voltage or PV module and transfer to laptop or
computer where the electrical signals and determinate the characteristic curve or PV
module.
5 Developing
All the analogies and integrated digital circuits, where signal transistors, resistances,
and capacitors are installed, were assembled by a printed circuit board. The power
transistor was installed on a heat sink because it works with 15 A of the current. The
control target was used to link the electronic circuit designed with the computer. Both,
electronics and control target are assembled into the cabinet. The electronic circuit
developed is shown in Fig. 5. Also, a cabinet was used for the installation of the
electric controls, as well as a digital display to show the values of current, voltage and
power that are utilizing to measure efficiency in the photovoltaic module. Photovoltaic
power meter prototype developed is shown in Fig. 6.
Fig. 5. Electronic circuit developed

Fig. 6. Photovoltaic power meter prototype developed.
6 Preliminaries Measurements
The maximum power, the I-V and P-V curves were measured with same level of
incident solar power and room temperature. Several photovoltaic modules of different
nominal powers were measured with the prototype developed. The measurements have
done in a normal day and the incident solar power was changing around from 600 to
860 W/m2. With the help of lux meter intensity (Solar power meter TES 1333R
Datalogging), two PV modules of 20 W manufactured in CINVESTAV Mexico were
used. The first measurement was done with one module and the second one was done
with two modules connected in parallel. The short circuit current (Isc), open circuit
voltage (Voc), maximum power (Pm), the maximum current (Im) and the Maximum
voltage (Vm) of photovoltaic module were measured. The experimental results are
shown in the Table 1.
Table 1. Shows the various parameters of PV module for the calculation of efficiency
PVm (W) Light power (W/m2) Iscc (A) Voc (V) Im (A) Vm (V) Pm (W)
20 600 1.12 18.45 0.91 12.8 11.52
20 700 1.24 18.51 1.15 12.00 14.04
20 860 1.25 18.64 0.91 12.7 17.24
40 790 2.31 18.75 2.15 12.9 27.55
40 800 2.36 18.60 2.24 13. 30.52
40 820 2.4 18.82 2.35 13.6 31.42
40 850 2.42 19.25 2.3 14.20 32.30
Also, the I-V curve was measured. The previous results are showed in the Fig. 7.
Here, the power voltage curve of PV module was plotted with the same incident
solar power was 700 W/m2. The previous results are showed in the Fig. 8.
Fig. 7. I-V curve of PV module measured
Fig. 8. P-V curve of PV module measured

Based on the experimental results of Table 1, it was found that the developed electronic
prototype works correctly and can measure the maximum power of the photovoltaic
module, and its current coordinates for varying levels of incident solar power. it is
possible to see how voltage and power variables of the photovoltaic module were
changed it was also confirmed by connecting two modules in parallel the current and
power variables increased while the voltage was almost the same value as that of a only
module.
In addition, it was observed that there was a greater change in current as a function
of the incident light power, however, the voltage change was minimal, which corre-
sponds to Eqs. 1 and 2 because the current changes exponentially while the voltage
changes logarithmically. From the voltage-current curve reported in Fig. 2, it could be
verified how the prototype developed can properly capture and process the current and
voltage signals of the photovoltaic module, store the information and then measure the
corresponding curve.
In the mentioned curve it was observed that the experimentally obtained curve
starts from zero with the horizontal axis of voltage this is due to the fact that the
measurement process of the photovoltaic module starts from the open circuit condition.
On the other hand the curve does not touch the vertical axis of current that is because
the electronic power circuit coupled to the photovoltaic module during the measure-
ment has a minimum resistance of the order of 0.3 V, which causes a voltage drop of
the order 0.6 V which prevents the curve from reaching the current axis. In relation to
the voltage power curve of Fig. 3, it is verified that the computer program developed in
Matlab which controls the measurement process works correctly because it is equal to
the theoretical curve of Fig. 1, which verifies that the voltage current processing was
right because it determines the power for all captured points.
However, it can be seen that the curve starts from the horizontal axis of voltage,
which indicates that the measurement starts in the open circuit condition of the pho-
tovoltaic module but does not reach the crossing point or zero of the axes, therefore it
reaches the short circuit condition, also due to the saturation of the power circuit of the
prototype developed. The development of a method of measuring the maximum power
of the photovoltaic module and its characteristic curves I-V and P-V propose is dif-
ferent of the proposed method. the purposed circuit is activated with a ramp voltage
signal in contrast with the reported methods which use technics like oscillating voltage
signal, pulse-width modulation, power signal derivation among others. Therefore, the
measurement with the proposed prototype is more stable and the proposed computer
program is simple. The proposed prototype can measure whatever photovoltaic module
with any incident light power.
8 Conclusions
According with the experimental results, the electronic system developed works cor-
rectly and its experimental results are agreement with the theoretical method used in the
design.
It is verified that the measurement technique of the photovoltaic module can be

controlled with a ramp voltage signal, and it is not necessary to use sophisticated and
complex methods such as those that have been reported. The electronic prototype has
an error of around 2%, but it is possible to improve it by selects precision electronics
devises and also by improving with the software. Although only 40 W power of
modules was measured in our experiment, it can measure up to 100 W. The prototype
was developed with electronics devices localized in Mexico; which is cheaper as well
as very easy to manufacture.
Acknowledgments. The authors are thankful to National Polytechnic Institute from México,
and also grateful to the project No. 20195869 SIP-IPN by their financial support.
References
1. Vega-Perez, J., Ponomaryov, V., Nino-de-Rivera, L.: Electronic system for power
optimization of photovoltaic generator. Int. J. Electromagn. Waves Electron. Syst. 7–8(8),
12–15 (2003)
2. Wanzeller, M., Cunha Alves, R., Fonseca Neto, J., dos Santos Fonseca, W.: Current control
loop for tracking of maximum power point supplied for photovoltaic array. IEEE Trans.
Instrum. Meas. 53(4), 1304–1310 (2004)
3. Munji, M., Okullo, W., Van Dyk, E., Voster, F.: Local devise parameter extraction of a
concentrator photovoltaic cell under solar spot illumination. Sol. Energy Mater. Sol. Cell 94,
2129–2136 (2010)
4. Ashish, P., Nivedita, D., Ashok, K.: High-performance algorithms for drift avoidance and
fast tracking in solar MPPT system. IEEE Trans. Energy Convers. 23(2), 681–689 (2008)
5. Khatib, T., Ghareeb, A., Tamimi, M., Jaber, M., Jaradat, S.: A new offline method for
extracting I-V characteristic curve for photovoltaic modules using artificial neural networks.
Sol. Energy 173, 462–469 (2018)
6. Danandeh, A., Mousavi, S.M.: Comparative and comprehensive review of maximum power
point tracking methods for PV cells. Renew. Sustain. Energy Rev. 82, 2743–2763 (2018)
7. Solimana, A., Hassana, H., Ookawara, S.: An experimental study of the performance of the
solar cell with heat sink cooling system. Energy Proc. 162, 127–135 (2019)
8. Peng, B., Chen, J., Liu, Y., Chiu, Y.: Comparison between three different types of variable
step-size P&O MPPT technique. In: International Conference on Computer Information
Systems and Industrial Applications (2015)
9. Santos, T., Galhardo, A., Perturbation, A.: Observation routine used to control a power
converter. In: 2014 Sixth World Congress on Nature and Biologically Inspired Computing
(NaBIC), pp. 78–83 (2014)
10. Farhat, S., Alaoui, R., Kahaji, A., Bouhouch, L.: Estimating the photovoltaic MPPT by
artificial neural network. In: Proceedings of International Renewable and Sustainable Energy
Conference (IRSEC), pp. 49–53 (2013)
11. Ramana, V., Jena, D.: Maximum power point tracking of PV array under nonuniform
irradiance using artificial neural network. In: Proceedings of IEEE International Conference
on Signal Processing, Informatics, Communication and Energy Systems (SPICES), pp. 1–5
(2015)
12. Wang, Y., Bian, N.: Research of MPPT control method based on PSO algorithm. In:
Proceedings of the 4th International Conference on Computer Science and Network
Technology (ICCSNT), pp. 698–701 (2015)
13. Badis, A., Mansouri, M., Sakly, A.: PSO and GA-based maximum power point tracking for
partially shaded photovoltaic systems. In: Proceedings of 2016 7th International Renewable
Energy Congress (IREC), pp. 1–6 (2016)
14. Rashid, M.: Power Electronics Handbook. Prentice Hall, Upper Saddle River (1999)
15. Malik, N.R.: Electronic Circuits: Analysis, Simulation, and Design. Prentice Hall, Upper
Saddle River (1998)
16. Sedra, A.S., Smith, K.C.: Microelectronics Circuits. Oxford University Press, Oxford (1999)
Artificial Intelligence and Machine
Learning
Clustering Methodology in Mixed Data Sets
Jacobo Gerardo González León(&) and Miguel Félix Mata Rivera(&)
Instituto Politécnico Nacional, Unidad Profesional Interdisciplinaria

en Ingeniería y Tecnologías Avanzadas, Avenida Instituto Politécnico Nacional
No. 2580, Col Barrio la Laguna Ticomán, Gustavo A. Madero,
jgonzalezl1007@alumno.ipn.mx, mmatar@ipn.mx
Abstract. One of the most challenging tasks of data analysis is finding clusters
in mixed data sets, as they have numerical and categorical variables, and lack a
labeled variable to serve as a guide. These clusters could serve to summarize all
the variables of a data set into one and be able to find information more easily
than generating summarizations for each variable. In this research thesis, a
methodology of clustering on mixed data sets is proposed, which yields better
results than the methods applied in the state of the art.
Keywords: Clustering Ensemble methods Mixed data set
1 Introduction
At today the human activity generates data every time, the problem is what to do with
them? How to find valuable information? As an interconnected technology society, we
already have stable data structures, industrial algorithms, repositories of almost all
topics, distributed computing to deal with large volumes but the problem remains, how
to process them?
This paper begins with the assumption of the existence of a hidden phenomenon in
these data sets that we will call a pattern [1]. The mathematic on which the data
analysis is based is focused on finding this phenomenon. Due to the complexity of the
high dimensions of the phenomena that describe our data sets, it is sometimes difficult
with simple statistics to find this pattern. That is why science never stops working and
keep looking for new ways to find information about the pattern hidden in our data
because no method has the absolute truth.
Then a specific methodology is proposed to process, through unsupervised clas-
sification, mixed data sets through a series of steps, which lead us to satisfactory results
as will be seen below, and which raises a new way of creating clusters beyond applying
algorithms directly. The steps are quantified with stable metrics, and especially this
methodology is intended to be usable in real problems.
The case study presented in this paper is the safe houses where the crime of
kidnapping was committed, the complexity here was to integrate and process different
mixed data sets from official surveys, to find valuable information that a specialist on
issues of public safety you may be interested.

https://doi.org/10.1007/978-3-030-33229-7_13
146 J. G. González León and M. F. Mata Rivera
In order to propose a complete and novel methodology, an exploration of works

related to “clustering for mixed data sets” in the formal scientific field in the 2013–2019
period corresponding to the 6 years prior to this proposal was made. The search term
was the clustering \ mixed data, and 89 papers were found, of which only 14 are
really related to the proposed work (10 algorithms, 2 methodologies and 2 imple-
mentations) because they meet the following criteria (see Table 1 where the works are
summarized):
• Type of work: the contribution is explicitly mentioned.
• Algorithm or proposed model: the contribution is explained.
• Mixed data set: the data set used, and the source are mentioned.
• Comparison: the contribution is compared against other similar works.
• Evaluation: the results obtained are measured.
Table 1. Summary of related research papers found in the state-of-art.

Type of work Algorithm or proposed Mixed data Evaluation
model set
Modified k-prototypes, FCM, UCI MLR Accuracy, F-measure,
algorithms Density Peaks Rand Index
Methodologies k-medoids, hierarchical Own case Dunn Index, Average
clustering study Silhouette Width
Implementation k-medoids Own case Average Silhouette Width
study
2 Unsupervised Classification in Mixed Data Sets

2.1 Mixed Data Sets
A mixed data set is a collection of X data represented in a structured tabular way
through an array of data of X rows, also called observations, instances, examples,
objects, points, tuples, etc., by D columns, also called attributes, properties, variables,
characteristics or dimensions and can be seen in Eq. 1 [2]:
0 1
X1 X2 ... XD
B x1 x11 x12 ... x1D C
B C
B x2 x21 x22 ... x2D C
X¼B C ð1Þ
B .. .. .. .. .. C
@ . . . . . A
xN xN1 xN2 ... xND
In where xi ¼ fx1i ; x2i ; . . .; xiD g are the observations, and Xj ¼ fX1 ; X2 ; . . .; XD g

are the variables.
The main characteristic of mixed data sets is that the domain of their variables can
take two types of values, numerical or categorical, also called quantitative and
Clustering Methodology in Mixed Data Sets 147
qualitative respectively; when a variable is numerical, its domain is of the real type,
therefore, its values can be continuous or discrete, finite or infinite. Its elements can be
part of any rational subset, integer, irrational or transcendent, then
domainðXnumerical Þ ¼ R; when a variable is categorical, its domain has a set of finite
symbols, its values can be numbers or characters that represent some category. These
categorical can be of two types; nominal, when there is no order in the symbols, and
ordinals, when, there is an order in the symbols, then
domain Xcategorical ¼ fs1 ; s2 ; . . .; sn g:
2.2 Unsupervised Classification

The human being when trying to describe a phenomenon generates a series of rules or
criteria, to understand the objects that describe its characteristics. Usually and given the
properties of a specific object to the phenomenon, classifying the information provided
by these objects, it may be useful to group these objects into categories to explain this
phenomenon [3].
Classification systems based on machine learning can be supervised or unsuper-
vised, it all depends on whether the characteristics of the object revolve around a
specific characteristic or attribute, called labeled variable, or there is a lack of this
variable, respectively. In the case of unsupervised classification, its objective is to
separate a set of data into groups, based on the concept of similarity or dissimilarity, as
the case may be, how similar or different are the objects?
This paper proposes the use of partition algorithms for unsupervised classification,
this is based on the optimization of an objective function that will iteratively improve
these partitions or clusters found. These algorithms have to be indicated the number of
partitions or groups in which you want to group the data.
3 Proposed Methodology
The most common way to attack the problem of unsupervised classification in mixed
data sets is through algorithms applied directly to the data, but in this research work,
the hypothesis is through a specific methodology for this type of data set (see Fig. 1),
could help in the search for clusters or groups in the data. Therefore, this thesis presents
a methodology to find groupings of observations in mixed data sets through the pre-
processing and application of an array arrangement of unsupervised partition classifi-
cation algorithms.
Steps 1 and 2, source and mixed data set, are subjective since the data set can come
from a specific experiment and the scientist who is analyzing the data, and they can
have an accurate knowledge of the phenomenon that is wanted to find or may also be in
exploration of the phenomenon.
Fig. 1. General diagram of the proposed methodology.
3.1 Preprocessing
This step consists of 4 sub-phases, load, clean, fill and transform, which have the
sequential sense to ensure the different actions: understanding the information, quality,
integrity and homogeneity in the data.
The mixed data set must be loaded into the computer’s memory to be able to
analyze it, and regardless of the operating system. The cleaning of a data set is done so
that the values correspond to the range of values that it should. One of the problems to
which this step gives solution is to deal with erroneous, inconsistent, obsolete and/or
redundant data. For this, it is proposed, with the help of metadata and/or prior
knowledge of the subject, to search and count the unique values of each variable and
compare the original domain with the current domain, in this way you can also easily
detect outliers and extremes
Once the anomalies in the data have been cleared, and if it is the case that there are
missing values, either spaces or any symbol that indicates a lack of data (NA, NaN, or
NULL) this missing data can be filled. Although there are several conventional
methods [4], among them, eliminating the rows that contain the missing values, or
replacing the missing value with the average, you can also use predictive methods [5]
based on supervised machine learning such as Linear Regression or Random Forest to
complete this lack of values.
The last sub-phase of the preprocessing is a transformation of the data, this in order
to normalize the domain in which the values of the variables are. As already mentioned
before, a mixed data set has numerical and categorical variables, for each type there is a
different transformation:
• Numerical: this transformation is made through the min-max normalization tech-
nique [6] with the minimum and maximum operations, respectively, which scales
all values to a range between [0, 1], according to the following Eq. 2:
xi minðXÞ
xi :¼ ð2Þ
maxðXÞ minðXÞ
In where X is the vector of the numerical variable to transform, and xi is the ith
value of vector X.
• Categorical: this transformation replaces the symbolic values of the categories of a
variable, the numerical values of its cumulative probability, and is based on the
proposal of [7] through the following method:
a. For each column of categorical data in the data set, the categories are sorted in
descending order by frequency.
b. Knowing the total observations, we find the probability pxi of each possible
categorical value xi through Eq. 3:
f xi
px i ¼ ð3Þ
N
In where xi is the ith value of the variable x; f xi is the frequency of observations of
the ith value of the variable x; and N is the total number of observations.
c. The categorical value is replaced by the value of its cumulative probability axi
through Eq. 4:
X
axi :¼ i
px i ð4Þ
3.2 Unsupervised Classification

This step consists of 6 sub-phases, clustering tendency, unsupervised classification
algorithms, internal evaluation, ensemble algorithm, consensus function, and final
partition. The internal evaluation sub-phases, unsupervised classification algorithms,
consensus function are found in an assembly structure, this structure combines different
algorithms and with a consensus function chooses the best partition within those
generated by the assembly.
Clustering Tendency
One of the first tasks when working with clustering data sets is to be able to measure if
it has relevant groups, that is, if there are the “grouping” conditions in the data set to be
analyzed. And although there is still no formal term, in [8] a “grouping function” is
defined as one that can quantify the degree of “grouping structures”, in other words,
find out whether or not the data set has groups significant.
Unsupervised Classification Algorithms
Beginning from the theorem “No free lunch” [9], which says that there is no “better”
machine learning algorithm that can solve a problem in a superior way because beyond
the proposed optimization function, it is true that Each set of data is unique in
dimensions, variables, and values. Then the grouping of the observations proposed in
this thesis is through an assembly of unsupervised partition classification algorithms
[10].
The unsupervised classification algorithms called “by partition” try to discover
groups in the data sets by optimizing a specific function and iteratively improving the
quality of these partitions [11] These algorithms require that they be previously indi-
cated the number k of groups, to begin measuring, the distance between observations
and these groups, is for this reason that this class of algorithms is also called “proto-
type” or “distance-based”.
In two or three dimensions there is not much difficulty in proposing the initial value
of the k clusters, because this parameter can be assigned visually, but in high
dimensions, it is difficult to define this value. That is why internal evaluation is pro-
posed, to find the best value objectively and measurably.
Internal Evaluation
Clustering is an unsupervised classification [12] that is, these algorithms attempt to find
an intrinsic structure in the data sets, and although there are different types of algo-
rithms of unsupervised classification, this structure commonly meets the similarity of
the data. This type of unsupervised classification techniques is applied to data sets with
no dependent variable, that is, they do not have a variable to predict.
In terms of accuracy and precision, and unlike the supervised classification, where
this dependent variable is used to assess how much the model or algorithm approxi-
mated the generalization of the data [13], in clustering it cannot be measured in the
same way, unless a priori the data set has the target variable and what you want, for
example, is to find perhaps what are the “true groups” or “true classification” [14] by
comparing the partition found against the classification original of the dependent
variable.
The evaluation of the found partitions of a mixed data set without labeling variable,
with unsupervised classification techniques, is called “internal evaluation” or “internal
measure” and is used to evaluate the quality of the partitions that are generated. Testing
different values of k [15] and measuring the quality of the partition, which found the
unsupervised classification algorithm that was used, through the similarity within the
groups and the separation between them, you can find this evaluation, commonly
through an internal evaluation index that is expressed by reason
ðintra-group)/(inter-group) [16]. For example, in [17] 27 internal evaluation indices are
proposed, whose value is the maximization or minimization of this ratio; the best
partition generated will then be the one that can have the least compactness within its
groups, and the highest among the separability among all.
Ensemble Algorithm
An ensemble of unsupervised classification algorithms and a heuristic that evaluate
different values of k and different unsupervised classification algorithms are proposed
in the following pseudocode:
Consensus Function
Once you have the validation matrix vm ½kn ; am ; pm ; em , a decision function f(vm Þ will
decide which is the best partition, by maximizing or minimizing the value of the
internal evaluation index em for each proposed group kn , in the following pseudocode
is showed a consensus function that maximizes:
3.3 Case Study

The phenomenon that was proposed to investigate to find groups of observations with
the proposed methodology for mixed data sets are: the safe houses where the crime of
kidnapping was committed. The data of these security houses were found using the
web scrapping extraction technique, searching for information in the following Mex-
ican news sites: Proceso, La Prensa, Reforma, and El Universal; extracting the fol-
lowing information: site, link, year, colony, municipality, and state. The address of
each safe house was found with the concatenation of the colony, municipality, and state
variables.
Once the address of the safe houses was extracted, a geocoding function was
applied to the address of each safe house to find its latitude and longitude. With 21
safety houses, a sample of 5 observations is presented in Table 2.
Table 2. Sample of 5 safe houses found.

Colony Municipality Latitude Longitude
Desarrollo Urbano Quetzalcóatl Iztapalapa 19.3263221 −99.050701
La Conchita Zapotitla Tláhuac 19.3068385 −99.043425
Casas Alemán Gustavo A. Madero 19.4783912 −99.089195
Manzanatitlán Cuajimalpa 19.367977 −99.299582
Agrícola Oriental Iztacalco 19.3956912 −99.088798
Figure 2 shows these security houses found on the map of Mexico City in the
2016–2017 period, in total 21 security houses were found distributed in 10 of the 16
municipalities.
The data found from the safe houses when being extracted from news do not come
with some extra information, beyond the address, that would allow us to be able to
identify the reason for the phenomenon, so it was devised to integrate different sets of
data that had characteristics of the geography where these houses were found, in order
to discover information.
Fig. 2. Map of the safe houses found in Mexico City.
According to the methodology developed in this thesis, these data must come from
an official source. Therefore, it was chosen to take open data from regular household
surveys of the Instituto Nacional de Estadística y Geografía (INEGI). Each of these
surveys is a set of mixed data, as there are questions with binary, or categorical
answers, as well as numerical data such as age, or amounts of money.
After completing the preprocessing phase proposed in the methodology, the mixed
data sets from the resulting surveys are shown in Table 3. Once the preprocessed mixed
data sets are done, in this case 5: HOGARES, SDEMT, TMOD, TPER_VIC1,
VIVIENDA, the unsupervised classification can start.
Table 3. Description of the variables of the surveys found.

Survey Mixed data set Numerical Categorical Observations
ENIGH Household information 10 65 1,733
(HOGARES)
ENOE Sociodemographic information 5 63 76,397
(SDEMT)
ENVIPE Victimization information 2 10 5,098
(TMOD_VIC)
ENVIPE Perception of security information 1 118 6,203
(TPER_VIC1)
ENH Housing information 8 53 3,588
(VIVIENDA)
3.4 Implementation
To begin with the measurement of the clustering tendency of mixed data sets, Hopkins
statistics were proposed [18] since this test that determines whether a data set has a
random structure, that is, measures the probability that the data set comes from a
uniform distribution of data. If the data sets do not come from a uniform distribution of
data, it may contain well segmented groups.
In [19] they review the unsupervised classification algorithms used over 20 years,
including partition algorithms, so it was proposed to use the k-means k-modes and k-
medoids algorithms. These algorithms have the particularity that the main parameter is
to indicate the k groups, so, according to this methodology, an internal evaluation index
is needed to find this number k.
The proposed internal evaluation index is the Dunn index [20] since it measures the
separation between groups, seeking the maximization of this inter-cluster distance, and
cohesion within the groups, seeking the minimization of intra-cluster distance. The
decision function will then be the maximization of this Dunn index [21] the larger the
Dunn index, the better the partition found. The index implementation could be bene-
ficial to find the optimal number of the k groups needed for the assembly algorithms
since this number of k could vary between 1, the case where the whole mixed data set is
a group, and N, where each observation is a different group.
4 Results
Earlier in the last chapter it was already mentioned, that the integration of information
to the case study “safe houses” is explored to describe this phenomenon, with an
unsupervised classification approach. It was also mentioned what is the source and
mixed data sets chosen for analysis, 5 surveys with different topics.
After processing the mixed data sets, the result of the clustering trend measurement
can be seen in Table 4.
Table 4. Result of the clustering tendency in mixed data sets from INEGI surveys.
Mixed data set Clustering tendency: Hopkins statistics
Without methodology With methodology
HOGARES 0.0461 0.3746
SDEMT 0.00814 0.2422
TMOD_VIC 0.0464 0.3759
TPERC_VIC1 0.1925 0.4908
VIVIENDA 0.1664 0.3552
Once it was measured how many significant groups, through the grouping trend,
can be found in these data sets, the unsupervised classification was performed with the
help of the algorithm assembly. Table 5 shows the result of the evaluation of the Dunn
index of the assembly of unsupervised classification algorithms proposed in the
methodology, against the algorithms that were found are used in the state-of-the-art k-
medoids and k-prototypes.
Table 5. Result of internal evaluation in mixed data sets from INEGI surveys.
Mixed data set Internal evaluation: Maximum Dunn Index
Without methodology With methodology
k-medoids k-prototypes k-means k-medoids k-modes
HOGARES 0.1040 1.94e−05 0.3872 0.0598 0.2326
Clusters 10 9 4 2 2
SDEMT 0.1810 7.0460e−06 0.6779 0.0515 0.6779
Clusters 2 2 2 10 2
TMOD_VIC 0.0271 0.0238 0.0334 0.0162 0.0331
Clusters 3 3 6 2 3
TPERC_VIC1 0.0078 0.0083 0.2458 0.1959 0.2884
Clusters 4 4 2 10 9
VIVIENDA 0.0819 0.0064 0.1892 0.1091 0.1302
Clusters 2 3 3 3 2
The unsupervised classification, through the algorithm assembly, proposed in the

methodology of this thesis, is designed to evaluate a set of mixed data at the same time.
The result of the unsupervised classification can be visualized the different groups
found with the help of the Principal Components Analysis (PCA) technique.
In Fig. 3 the different groups can be displayed, as well as the map of Mexico City,
the delegations belonging to each group found from the HOGARES survey, and
Table 6 shows a sample of 5 variables that characterize each group found.
Fig. 3. Visualization of the groups found of the mixed data set HOGARES.
Table 6. Sample of 5 variables of the groups found in the mixed data set HOGARES.
Survey: HOGARES
Variables acc_alim1 acc_alim3 acc_alim6 alim17_2 alim17_4
Cluster Meaning
Worry Limited Number of Number of Number of
meal is variety of days they ate days they ate days they ate
over food egg meat fruits
1 No No 2 2 7
2 Yes Yes 3 2 2
3 No No 2 7 7
4 No No 3 3 7
the delegations belonging to each group found from the SDEMT survey, and Table 7
shows a sample of 5 variables that characterize each group found.
Fig. 4. Visualization of the groups found of the mixed data set SDEMT.
Table 7. Sample of 5 variables of the groups found in the mixed data set SDEMT.
Survey: SDEMT
Variables Sex eda cs_p13_1 cs_p17 e_con
Cluster Meaning
Sex Age What degree did the Do you currently Marital
school approve? attend school? status
1 Woman 21 Elementary Yes Single
2 Man 40 Undergraduate No Married
the delegations belonging to each group found from the TMOD_VIC survey, and
Fig. 5. Visualization of the groups found of the mixed data set TMOD_VIC.
Table 8. Sample of 3 variables of the groups found in the mixed data set TMOD_VIC.
Survey: TMOD_VIC
Variables BP1_1 BP1_5 BP1_3C
Cluster Meaning
What month did the Where place did the In which municipality did
crime happen? crime happen? the crime occur?
1 May On the street Álvaro Obregón
2 February On the street Iztapalapa
3 December On the street Iztapalapa
4 December In his house Álvaro Obregón
5 August On the street Benito Juárez
6 December In his house Cuauhtémoc
the delegations belonging to each group found from the TPER_VIC1 survey, and
the delegations belonging to each group found from the VIVIENDAS survey, and
Fig. 6. Visualization of the groups found of the mixed data set TPER_VIC1.
Table 9. Sample of 3 variables of the groups found in the mixed data set TPER_VIC1.
Survey: TPER_VIC1
Variables AP4_10_01 AP4_10_04 AP4_10_06
Cluster Meaning
For fear of being a victim For fear of being a For fear of being a
of a crime, did you stop victim of a crime, did victim of a crime, did
going out at night? you stop taking a taxi? you stop carrying cash?
1 Yes No Yes
2 No No No
3 Yes No Yes
4 No No No
5 No No Yes
6 No No No
7 No No Yes
8 No No No
9 Yes Yes Yes
Fig. 7. Visualization of the groups found of the mixed data set VIVIENDAS.
Table 10. Sample of 5 variables of the groups found in the mixed data set VIVIENDAS.
Survey: VIVIENDAS
Variables antiguedad tipo_viv num_cuarto cuart_dorm auto_pick
Cluster Meaning
Housing Type of Number of Bedroom Own car
age housing rooms availability
1 27 Independent 5 2 Yes
house
2 25 Independent 3 1 No
house
3 32 Department 4 2 Yes
5 Discussion
With only the information from the mixed data sets from the INEGI surveys in Table 3,
we could begin to generate hypotheses about how many groups these data sets could
have. For example, from the SDEMT table that contains more observations, we might
think that there are many more groups than the others, or that the HOGARES table that
contains fewer observations, there are few groups. So, when we apply the methodol-
ogy, we realize that only two groups were found in the SDEMT survey, while four
groups were found in the HOGARES survey. There is not yet a rule that, depending on
the number of observations or variables, so many groups will be obtained, which is
why the unsupervised classification exists and makes sense, to propose this number of
partitions in the data that at first glance cannot be defined. Also, with the basic
information of the groups for each mixed data set, summaries could be made with their
variables. but summary tables should be made for each variable that contains the mixed
data set, this can take a long time and many tables, so it might not be useful.
One advantage of finding clusters in the data is to simplify this task to only one
table, as for example, in the case of Table 6 of the first HOGARES survey analyzed. It
is the case similar to wanting to see on a map the characteristics of each delegation, the
same number of maps would have to be made to be able to visualize all the variables, as
is the case of Fig. 3 of the HOGARES survey, where the map is seen with the dele-
gations belonging to each group found, it is easier to read and interpret the variables of
a mixed data set, based on the groups found with this methodology.
Then the interpretation for each of the data sets, based on the question: which
delegations do not have safe houses? You can respond by summarizing the groups
found. Even with the methodology supported by the different measurements that are
made, questions remain that perhaps only an expert on the subject could answer.
Questions like if the groups found have any interest value? Or in what range of k
clusters should the heuristics of the algorithm assembly be run?
5.1 Conclusion and Future Work

This work is aimed to improve the search for groups in mixed data sets based on a
review of the state of the art, to see how the unsupervised classification of their
observations was being done. A methodology was then proposed with a series of steps
that show a better generation of groups, than simply applying an algorithm to these data
sets. The results obtained from the implementation of this methodology, in the case of
study and the extension, are satisfactory and show that they improve the creation of
groups.
To find the best partition, or the best groups, Hopkins’s statistics were proposed so
that before processing, one could know what the distribution of the data set is like,
because if it has a uniform nature, few groups could be found, and if it has a non-
uniform nature, significant groups could be found. This can be seen in the case of the
SDEMT survey, which was the least valuable data set, almost 0 and only 2 groups were
found; contrary to the TPERC_VIC1 survey, which was the most valuable data set,
almost 0.5 and 9 groups were found. This number of clusters was found with the
maximization of the Dunn index, which, the larger, is interpreted as which better
groups have been created.
As future work, more internal validation rates, other measures to evaluate the
clustering tendency, could be integrated, as well as other real mixed data sets to
continue measuring the effectiveness of this methodology.
References
1. Ströing, P.: Scientific Phenomena and Patterns in Data. Ludwig-Maximilians-Universität,
München (2018)
2. Zaki, M.J., Meira, W.: Data Mining and Analysis. Cambridge University Press, Cambridge
(2014)
3. Bramer, M.: Principles of Data Mining. Springer, London (2016). https://doi.org/10.1007/
978-1-4471-7307-6
4. Soley-Bori, M.: Dealing with missing data: key assumptions and methods for applied
analysis, vol. 23. Boston University (2013)
5. Yadav, M., Roychoudhury, B.: Handling missing values: a study of popular imputation
packages in R. Knowl.-Based Syst. 160, 104–118 (2018)
6. Larose, D., Larose, C.: Discovering Knowledge in Data: An Introduction to Data Mining,
2nd edn. Wiley, Hoboken (2014)
7. Patki, N., Wedge, R., Veeramachaneni, K.: The synthetic data vault. In: IEEE International
Conference on Data Science and Advanced Analytics (DSAA), pp. 399–410. IEEE (2016)
8. Adolfsson, A., Ackerman, M., Brownstein, N.: To cluster, or not to cluster: an analysis of
clusterability methods. Pattern Recogn. 88, 13–26 (2019)
9. McCue, C.: Public-safety-specific evaluation. In: Data Mining and Predictive Analysis:
Intelligence Gathering and Crime Analysis, pp. 157–183. Butterworth-Heinemann (2015)
10. Wu, X., Ma, T., Cao, J., Tian, Y., Alabdulkarim, A.: A comparative study of clustering
ensemble algorithms. Comput. Electr. Eng. 68, 603–615 (2018)
11. Jukes, E.: Encyclopedia of machine learning and data mining (2nd edition). Ref. Rev. 32, 3–
4 (2018)
12. Loshin, D.: Knowledge discovery and data mining for predictive analytics. In: Business
Intelligence. The Savvy Manager’s Guide MK Series on Business Intelligence, 2nd edn.,
pp. 271–286 (2013)
13. Tharwat, A.: Classification assessment methods. Appl. Comput. Inform. (2018)
14. Hennig, C.: What are the true clusters? Pattern Recogn. Lett. 64, 53–62 (2015)
15. Gurrutxaga, I., Muguerza, J., Arbelaitz, O., Pérez, J., Martín, J.: Towards a standard
methodology to evaluate internal cluster validity indices. Pattern Recogn. Lett. 32, 505–515
(2011)
16. Jauhiainen, J., Kärkkäinen, S.: Comparison of internal clustering validation indices for
prototype-based clustering. Algorithms 10, 105 (2017)
17. Desgraupes, B.: Clustering Indices. University of Paris Ouest-Lab Modal’X, vol. 1, pp. 34
(2013)
18. Han, J., Kamber, M., Pei, J.: Cluster analysis: basic concepts and methods. In: Data Mining,
pp. 443–495 (2012)
19. Benabdellah, A., Benghabrit, A., Bouhaddou, I.: A survey of clustering algorithms for an
industrial context. Proc. Comput. Sci. 148, 291–302 (2019)
20. Rodriguez, M., Comin, C., Casanova, D., Bruno, O., Amancio, D., Costa, L., Rodrigues, F.:
Clustering algorithms: a comparative approach. PLoS One 14, e0210236 (2019)
21. Yang, Y.: Temporal Data Mining via Unsupervised Ensemble Learning. Elsevier Science,
Amsterdam (2016)
American Sign Language
Electromiographic Alphabet
Sign Translator
Edgar-Armando Catalan-Salgado(B) , Cristhian Lopez-Ramirez,

and Roberto Zagal-Flores
Instituto Politecnico Nacional (IPN), Escuela Superior de Cómputo (ESCOM),

Av. Juan de Dios Batiz s/n, GAM, Mexico City, Mexico
{ecatalans,rzagalf}@ipn.mx, cristhlopram@gmail.com
Abstract. Communication between people is complicated, thus is due

to correct idea and thought expression. But for the deaf or mute people
this is even worse due to that our main communication channel is sound.
They can use their own language using sign and ideograms made with
hands, called American Sign Language. But as every language it is needed
to learn and the population that dominate this language is small. In this
work we propose an American Sign Language translator for 24 alphabet
signs, using a wearable that give us eight electromiographic signals and
KNN classifier for signs processing with 80% of accuracy.
Keywords: American Sign Language · KNN · Pattern recognition ·

Artificial intelligence and health
1 Introduction
Communication is in some way complicated between different beings. Correct

ideas, thoughts and desires transmission is not ever correctly or fully realized.
But this is even harder for impaired people, particularly deaf or mute people,
this is due to the fact that spoken language is our main communication language.
Disability is a result between people with impairments and the barriers that
arises due to attitudes and environment that restrain them for the fully par-
ticipation in society [7,8]. In Mexico, this definition is specified for people with
some physical, mental, intellectual or sensorial deficiency [5].
Already exists a variety of languages that only uses hands signs and move-
ment to establish communication, one of them is called American Sign Language
(ASL) [1], is mainly used by deaf or mute people and his family or near friends.
But find other people that dominates this language can be a challenge.
In this paper we establish a novel method to translate the signs correspond-
ing to the ASL alphabet to text, in order to give the opportunity to establish
communication for impaired people.

https://doi.org/10.1007/978-3-030-33229-7_14
ASL Electromiographic Sign Translator 163
Automatic sign recognition is a well established problem and exist differ-

ent approaches in literature. First of all, it is necessary to consider the gesture
components such as hand shape, location and movement [2].
Many artificial vision systems have been developed in [2,9–11], nevertheless
they are based on a camera using and lighting dependency. In other words, their
performance is affected on situation where there is high or a low illumination. For
example, when a leap motion controller (LMC) is used that uses a combination
of 2 cameras to make a three-dimensional reproduction of hands and their space
movement [3], using this sensor and considering 26 alphabet letter, Chang et al.
produces 80.30% of accuracy using a Support Vector Machine (SVN), but using
a deep neural network (DNN) a 93.81% of performance is developed.
An alternative to vision artificial systems is the use of forearms wearables
and uses electromiographic and motion sensors. An example of this is the Myo
Arm band, which one deliver Inertial and Electromiographic (EMG) data. This
kind of wearable approach have already been used in previous research, partic-
ulary Fatmi et al. [4] used two myo arm band devices, one for each forearm,
and considering the IMU and EMG data as time series tested Artificial Neural
Networks (ANN), Support Vector Machines (SVM) and Hidden Markov Models
(HMM) with a mean accuracy of 93.79, 85.56, and 85.9 respectively.
We used as input eight electromiographic signals, given for EMG sensors of
a Myo Arm Band in the right forearm person. Each one signal was preprocessed
getting a muscle activation parameter. Using the eight activation parameter as
feature vector for each sign a KNN classifier was used. Experimentation using
24 signs of ASL alphabet deliver a 80% accuracy.
2 American Sign Language

A sign language uses visual-manual modality to convey meaning [14]. It is
expressed through manual articulations and full fledged natural languages with
its own grammar and lexicon [12]. Wherever a deaf people exist a version of sign
language arises. The 2013 edition of Ethnologue list 137 sign languages [6].
American Sign Language (ASL) is an the predominant sign language used
mainly by deaf and mute community in United Sates and canada [13], with
a user number range between 250,000 and 500,000 persons. It consist in signs
and ideograms, made with hands and specific movements that transfer a specific
letter or idea. ASL possesses 26 signs know as the American Manual Alphabet
that are shown in Fig. 1. Fingerspelling or dactilology is the representation of the
letters of a writing system and numeral systems using only the hands. In ASL,
fingerspelling is used for proper nouns, abbreviations and for technical terms
with no native ASL equivalent.
3 Electromiographic Sensors
Electromyography (EMG) and nerve conduction study (NCS), are tests that
use electrodes to detect, translate, and record the electrical signals in muscles
164 E.-A. Catalan-Salgado et al.
Fig. 1. American Sign Language Alphabet
and nerve cells, generally used by specialist in order to determine the causes
of diseases that affect muscles and nerves, but lately have been used to make
automatic ASL translation.
Between the commercial grade devices, the Myo Arm band excels over others,
it have a EMG sensors and Inertial Movement Unit (IMU), at a relative low price,
the wearable is shown in Fig. 2.
Fig. 2. Myo arm band
Data collected is transmitted to a compatible device using bluetooth, IMU

data at 50 Hz and EMG data at 200 Hz. In synthesis, IMU data collect infor-
mation from a three axis giroscope, three axis magnetometer and three axis
accelerometer; EMG collect information of eight EMG sensors, delivered as a

byte array of eight elements whose values are between −128 and 127 and unit-
less. So trough a time the signals would be as shown in Fig. 3.
Fig. 3. Sensors values in time
4 Method
When we made a sign, the position of our fingers, hand and wrist changes. Thus
undoubtedly changes the tension of our forearm muscles, changing also the emg
signals. So theoretically the combination of the eight signal values in time t can
represent a sign, but can’t be directly used due to their oscillating nature, so a
processing is needed in order to be useful as a muscle activation measure.
As a muscle activation measure for a given time t, a standard deviation σ of
each sensor s is calculated for the period [t−n, t], where n is a period size. This
is calculated as follows:
t

x̄st = (1)
i=t−n

t s
i=t−n (xi− x̄st )2
σts = (2)
n−1
With this we can get the following row vector in a given time t:

Xt = σt1 , σt2 , σt3 , σt4 , σt5 , σt6 , σt7 , σt8 (3)
For sign recognition purposes the time is not needed, so we can take it out
from vector showed in 3, and let it as follows:

X = σ1 , σ2 , σ3 , σ4 , σ5 , σ6 , σ7 , σ8 (4)
This vector can be interpreted as a measure of the forearm muscles activation

detected by each sensor s. Graphically this abstraction can be observed in Fig. 4.
Some things worthy to remark are:
First, the mean value is almost zero for every sensor.
Second, the range values is not the same in this case for some is [−10, 10], for
others [−25, 25] and for others [−50, 50], but according to technic specifications
could be any possible between [−127, 127].
Third, exist a correlation between value range of each sensor, and the muscles
used to make a sign.
Finally, by definition, high variations in considered window time would bring
high values in standard deviations values.
From previous points our theory is that we can use standard deviation values
as a measure of muscle activation in a given window time, and used it to classify
hand signs.
Fig. 4. Standard deviation in a range [t−n, t]
4.1 Classifier
For simplicity a KNN classifier was used for testing, due to is a easy algorithm
and doesn’t have parameters that can alter or bias to the final classification,
allowing a general behaviour. In this case, we used the value 1 for K parameter.
4.2 Learning Phase

The number of signs taking in account is 24. Then, an index g∃[1, 24] is defined
and each sign can be addressed by g. The KNN clustering algorithm is used for
one sample per sign, in order to create a matrix of 24 rows and 8 columns. This
strategy allows to simplify the learning phase.
The learning algorithm is defined as follows:
1. Get X vector for each sign g.

Xg = σg1 , σg2 , σg3 , σg4 , σg5 , σg6 , σg7 , σg8 (5)
2. Introducing Xg in KNN matrix:
mg,i = xig (6)
In this way the matrix M becomes in:

⎡ 1 ⎤
σ1 , σ12 , σ13 , σ14 , σ15 , σ16 , σ17 , σ18
⎢ ... ⎥
⎢ 1 ⎥
M =⎢ σ
⎢ g , σ 2
g , σ 3
g , σ 4
g , σ 5
g , σ 6
g , σ 7
g , σ 8⎥
g⎥ (7)
⎣ ... ⎦
1 2 3 4 5 6 7 8
σ24 , σ24 , σ24 , σ24 , σ24 , σ24 , σ24 , σ24
4.3 Recall Phase

A euclidean distance d is used to recall, due to the fact that for each sign g only
exists one input in the KNN matrix M, then a 1-NN classifier was chosen.
The recall algorithm is:
1. Get the X vector for the unknown sign

X = σ1 , σ2 , σ3 , σ4 , σ5 , σ6 , σ7 , σ8 (8)
2. Get the euclidean distance to each one of the learned patterns.

8
2
d(x, mg ) = xi − mig (9)
i=1
3. Assign as output class those whose d(x, mg ) is the minimum.
5 Experiments and Results

The signs corresponding to the letters j and z was excluded due that they need
some hand movement. Also the numbers was excluded and would be considered
in future work. So the remaining 24 signs in the ASL alphabet are the letters
that the classifier can recognize as the possible output classes.
First of all a time window n = 20 is considered in order to get the standard

deviations.
For training phase the X vector of standard deviation was obtained for each
one of the 24 sign considered and learned with the KNN.
For recall phase ten different instances of each sign from the same user is
considered. For each one the X vector of standard deviations is obtained and
tested with with the KNN.
As validating method a confusion matrix is used, this is due to additional
information about the algorithm behaviour that can be achieved by this matrix.
Due to the matrix confusion size this was divided in three parts of seven signs
each one, showed in Tables 1, 2 and 3.
Table 1. Confusion Matrix part 1, a–h
X a b c d e f g h i k l m n o p q r s t u v w x y %
a 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100
b 0 4 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 40
c 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 20
d 0 0 0 5 0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 50
e 0 0 0 0 9 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 90
f 0 6 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 40
g 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100
h 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100
Table 2. Confusion Matrix part 2, i–q
i 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100
k 0 0 0 0 2 0 0 0 0 0 0 0 0 2 0 0 6 0 0 0 0 0 0 0 0
l 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 0 0 100
m 0 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 0 2 0 0 0 0 0 0 80
n 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 100
o 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 100
p 0 0 0 0 0 0 1 0 0 0 1 0 0 0 6 2 0 0 0 0 0 0 0 0 60
q 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 9 0 0 0 0 0 0 0 0 90
This gave us a global performance of 80%, that is a high percentage consid-

ering the number of classes.
Table 3. Confusion Matrix part 3, r–y
r 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 100
s 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 7 3 0 0 0 0 0 70
t 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 9 0 0 0 0 0 90
u 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 100
v 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 100
w 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 9 0 0 90
x 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0 100
y 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 100

According to the result the measure of muscle activation is a good discrimi-
nant for hand sign recognition, using a KNN classifier with 80% of efficiency is
achieved. As a result, it becomes an excellent feature, because our approach uses
low computational and spacial cost, this advantage is very important when the
recognition algorithm it is necessary to be implemented in a limited hardware
capabilities such as mobile devices, wearables or other small devices, allowing the
use of low-computational and low-space cost algorithm like Nearest neighbour
algorithm (1NN). In contrast, more complex algorithms like ANN or support
vector machines would take advantage of this measure and achieve even better
results.
Although comparison with other algorithms is desired, there isn’t a basic data
set, an alternative option is to choose ideograms randomly but any comparison
isn’t fairly. In this paper the 24 static alphabet sign was considered as a data set,
while others authors such Fatmi et al. [4] used only 13 signs obtaining 82.91% of
accuracy as minimum, but the ideograms considered both hands and movement,
it could be a disadvantage.
The most similar dataset in literature is the used by Chong et al. [3], they
used the 26 alphabet sign letters, achieving a 80.30% accuracy using a Support
Vector Machine and 93.81% using a Deep Neural Networks (DNN). In the case
of SVM is almost the same accuracy, but using DNN they won by at least 14%,
however both algorithms have high computational cost.
Making a deeper analysis of confusing b with f, and k with r, because hands
signs are very similar and only change orientation. Others would need more
analysis in a different levels like the worst result that is confusing c and v and
doesn’t seem to be a logical reason and the proposed measure fails.
As future work, the reduction of this confusion and improvement of the clas-
sification is desired, in order to use additional features or another classifier that
brings weight to features, maybe an Artificial Neural Network. Another impor-
tant task is to consider the hand movement analysis for not only to include the
j and z letters based on ideograms recognition, also the analysis the Inertial
Movement Unit of the Myo Arm Band. Another way to improve this research is
the use of two arm band, one for each forearm, in order to be able to recognize
ideograms made with two hands. In fact, the research of low cost features would
benefit all the sign recognition disciplines.
Acknowledgments. The authors are grateful to Instituto Politecnico Nacional for

the economic support given to the research project number 20196065 given through
the Secretaria de investigacion y posgrado.
References
1. ASL: Ethnologue Definition, American Sign Language (ASL) (2019). https://www.
ethnologue.com/language/ase
2. Banerjee, A.E.A.: Generation of movement explanations for testing gesture based
co-operative learning applications. In: Proceedings - 2019 IEEE International Con-
ference on Artificial Intelligence Testing (2019)
3. Chong, T.W., Lee, B.G.: American sign language recognition using leap motion
controller with machine learning approach. Sensors (Switzerland) 18, 3554 (2018)
4. Fatmi, R., Rashad, S., Integlia, R.: Comparing ANN, SVM, and HMM based
machine learning methods for American sign language recognition using wear-
able motion sensors. In: 2019 IEEE 9th Annual Computing and Communication
Workshop and Conference (2019)
5. INEGI: Disability in Mexico. INEGI (2014)
6. Lewis, M.P., Simons, G.F., Fennig, C.D.: Ethnologue: Languages of the World,
17th edn. SIL International, Dallas (2013)
7. OMS: World report on disability. OMS (2014)
8. ONU: impaired people rights conference. ONU (2017)
9. Pansare J.R., Ingle, M.: Vision-based approach for American sign language recogni-
tion using edge orientation histogram. In: 2016 International Conference on Image,
Vision and Computing, ICIVC 2016 (2016)
10. Torung V.N.T., Yang, C.K., Tran, Q.: A translator for American sign language to
text and speech. In: 2016 IEEE 5th Global Conference on Consumer Electronics,
GCCE 2016 (2016)
11. Vanco, M., Minarik, I., Rozinaj, G.: Evaluation of static hand gesture algorithms.
In: International Conference on Systems, Signals, and Image Processing (2014)
12. Sandler, W., Lillo-Martin, D.: Sign Language and Linguistic Universals. Cambridge
University Press, Cambridge (2006)
13. Wikipedia: American Sign Language. https://en.wikipedia.org/wiki/American
Sign Language
14. Wikipedia: Sign Language. https://en.wikipedia.org/wiki/Sign language
Artificial Intelligence as a Competitive
Advantage in the Manufacturing Area
Juvenal Mendoza Valencia(B) , Juan José Hurtado Moreno,

and Felipe de Jesús Nieto Sánchez
Instituto Politécnico Nacional, UPIITA, G.A. Madero, 07340 Mexico City, Mexico
juvenalmv69@gmail.com, hurtadoupiicsa@yahoo.com, felnieto73@gmail.com
Abstract. Since the beginning of the industrial revolution, manufactur-

ing has gone through different stages: the 1st technological islands, 2nd
the mass production, 3rd the lean manufacturing and the 4th IIoT (for
the year 2025); we must keep in mind the leadership in the production
of goods today what the Eastern countries have (and are using stage 3),
so that the current guidelines have the necessary meaning, which estab-
lishes a new way of producing more as the potential of technologies that
are in the process of maturation such as: Artificial Intelligence, Big Data,
3D Printing and Robotics; find original solutions to the problems of pro-
ductivity, customization, just in time and services.
Artificial Intelligence is taking a leading role in solving manufactur-
ing problems, with the purpose of eliminating all those areas that are
blindly worked, and that therefore it is not possible to improve by suf-
fering from data to: analyze them, obtain information, establish controls
and improve. The above is achieved by establishing disruptive technolo-
gies that take control in real time.
Keywords: Industry 4.0 · Artificial Intelligence · Big data · 3D

Printing and Robotics
1 Introduction
The Oriental companies that apply the Lean concept (3rd stage), have two funda-
mental guidelines and without them it cannot operate properly [11]. The First
one, a hierarchical structure is horizontal (democratic), which implies that all
the members of the company have administrative and operational functions, so
that decision-making is the result of a consensus.
The Second one, its based on first sell and then produce, that means,
what the company is looking for is not to work blindly, the programming of
the orders is the result of real orders; for this purpose it is necessary to bring
transparency to the entire production process, making visible the stoppages in
real time; achieving this comply with the agreed with customers.
Western companies [11]. Been preferred to maintain a vertical structure
(Authoritarian), so the work environment is very stressful and they seek to pro-
duce and then sell; but when the first premise is not fulfilled, it is generated that
https://doi.org/10.1007/978-3-030-33229-7_15
172 J. Mendoza Valencia et al.
the established work program is not achieved, generating delays, poor quality
products and penalties; the above is the result of the low participation of workers
in making decisions, as well as excessive rotation of them at different levels.
In the different research works in manufacturing companies in Mexico, it
has been found that their managers are more focused on achieving short-term
results, such as compliance with the volume of production, leaving aside quality,
variety and time of delivery.
The problem is that managers do not understand the philosophy of lean
manufacturing, so by not complying with the premises indicated above, their
companies do not have the possibility of joining a global production chain, which
requires them to comply with the times of delivery, international prices and
quality of its products.
As companies of Japanese origin have arrived, they have worked with their
suppliers in the teaching and understanding of their work methods, which has
allowed a greater number of CEO, managers and workers to understand these
forms of production.
25 years have passed since the signing of the free trade agreement between
the United States, Canada and Mexico, it is expected that in this second stage
governments will require their companies as a requirement to operate the imple-
mentation of Lean Manufacturing, which will put their companies of manufac-
turing at the same level as the Oriental companies.
With the above [10], the foundations are laid for equitable competition
between the different countries, which means that they are able to move to indus-
try 4.0. They have 5 years for the installation of 5G communication technology,
which will allow the development of the IIoT industry (Internet of industrial
things).
1.1 4.0 Industry Considerations

Lean Manufacturing was born from the idea of controlling production through
the computer [3], for this purpose the Kan-Ban was created, this is nothing
more than a set of cards that were taken to the computerized processing center
in periods of 2 h and In this way it was possible to know if the volume of goods
corresponded to what was programmed; from this restriction the requirement
arises that all the flow of internal and external materials had to be done in lots
that will not exceed that time; By involving all areas of the company, there was
a need to develop a set of methodologies to solve the different manufacturing
problems.
The use of the computer in the 60s represented a great innovation in the
production processes and this did not stop there, also followed developments such
as the use of robots, networks and communication equipment, CNC machines,
PLCs, ERP Software (planning of the business requirement), etc. All these new
technologies were introduced harmoniously in the industry and helped each other
to increase its productivity, quality and flexibility.
Each new technological development that was introduced in the industry
generated changes in the way of producing, resulting in the elimination of jobs,
Artificial Intelligence as a Competitive Advantage in the Manufacturing Area 173
which were rewarded by a considerable increase in productivity, as well as the

profits of the company, the above made that it grows and demands workers in
areas such as: technical, administrative, logistics, marketing, etc., resulting in a
benefit of the economy as a whole.
We have many disruptive technologies in the 4.0 industry [2], each one solv-
ing a specific problem, we would have to ask, which of these has a strategic
importance in the production of goods?; The answer is Artificial Intelligence, it
allows machines to do their jobs better and improve in time, never in the history
of mankind have we transferred intelligence to them, since the beginning of the
industrial revolution equipments have been built, which they follow a set of steps
in a mechanical, electrical or electronic way, which has allowed the human being
to be released from monotonous and dangerous work.
With the creation of the smartphones in 2007 [10], companies have developed
a set of smart devices and softwares, as a way to differentiate themselves from
their competitors, in many cases it has represented a competitive advantage,
which forces those who do not to imitate these, as a way to remain in the
market.
We must consider Artificial Intelligence [2], as a new element of integration
of the different areas of the company, being part of the ERP system, whose
main function is to take control of production in real time with the advantage
of knowing all the productive incidents at throughout the entire production
chain; quantifying therefore all the delays that occur in the same, presenting in
a graphic way the state that the different orders keep “e.g.” (red delay of more
than 2 h, Yellow of 1 h and green without delays).
We must consider Artificial Intelligence as a new element of integration of the
different areas of the company, being part of the ERP system, whose main func-
tion is to take control of production in real time with the advantage of knowing
all the productive incidents at throughout the entire production chain; quantify-
ing therefore all the delays that occur in the same, presenting in a graphic way
the state that the different orders keep “e.g” (red delay of more than 2 h, Yellow
of 1 h and green without delays). To this day it is how control of the manufactur-
ing plants is carried out; but in reality the use of the different applications that
make up this platform is between 50 to 80%, this is due to the problem of trans-
ferring information from different because of geographical issues and productive
conditions, thus leaving many areas that work blindly.
2 Methodology
If we consider that Western companies have a Vertical Organization, it does not

generate an effective participation of workers, they are only limited to following
orders, without analyzing whether these are adequate to the needs of the partic-
ular production of the company, seeing the results of these decisions at the end
of the fiscal year.
It is evident that all this generates a poor work culture and a lack of interest
of workers to solve everyday problems, because they are excluded as an element
that can bring creative ideas to the company, If this is taken into account, it
will substantially change the way it operates, being part of the solution of the
production process.
2.1 Inefficiency of Equipment and Staff Turnover

We must have defined where our strengths and weaknesses are, they will make
us aware of the importance of technology to the achievement of the development
of a new competitive strategy, as well as the elimination of our inefficient areas,
One of the most frequent problems that manufacturing companies have is the
separation of the master production program and the actual execution of the
same, due to the fact that they have countless number of machine tools, these
occur when operating endless of failures, which have to be repaired at the time
of production, which generates a cumulative amount of time that consumes more
than scheduled.
For the measurement of the problem of failures, the most appropriate solution
so far, is to apply the measure called: General Equipment Efficiency (OEE),
which establishes that for world-class companies you must have an efficiency of
85% [9], therefore there is a 15% for the correction of machine failures, if it will
work optimally we would only charge an over 15% cost and delivery times would
be those involved when considering this percentage of inefficiency at the time of
scheduling production; But for this to be true, we must have low staff turnover
and that it perform decision-making functions, which does not occur in most of
Western companies.
To reduce staff turnover, it is an essential requirement to have a minimum
level of technological baccalaureate studies, which allows workers to under-
stand the potential of new technologies and therefore make contributions to
the improvement of processes in an innovative way, achieving development and
job satisfaction, creating a healthy work culture.
If this is fulfilled in terms of complying with the specifications of the equip-
ment failures, minimum academic training and job satisfaction, we will be able
to use disruptive technologies to build a competitive advantage.
2.2 Artificial Intelligence and Disruptive Technologies that Support

Production
If we apply [7] Artificial Intelligence, intelligent machines, robots, products with
embedded systems with Radio Frequency Identification (RFID) and digital twins
companies, would be able to receive production orders in real time, analyze the
financial part, define time of delivery, track orders, analyze delivery logistics, see
how the products work with the customer, update the product software offline,
make fault diagnostics and define a solution.
It is important to explain how these technologies would interrelate, [15] in
the first place, the function of Artificial Intelligence (AI), is to give intelligence
to things, for this purpose it is necessary that they have an embedded system,
which monitors the actuators and motors For this purpose, sensors must be
included: temperature, position, speed, geolocation, touch, visualization, listen-

ing and speaking, with which it is possible to communicate directly between
the manufacturer and the user, to indicate the required maintenance to indicate
the changes of the parts, the time of life and define the date of replacement of
the device; It would increase the productivity of the company with the use of
this new technology (AI), being able to analyze data and make decisions before
the problem of equipment failures occurs, thus achieving a general efficiency
of the 100% equipment (OEE), and therefore eliminating the 15% cost over-
run, which had to be considered before, when they worked without an effective
communication.
What we seek with the technologies of Industry 4.0 is to move from static
to dynamic systems [13] (that is, they can be changed over time), the reason
is that markets are unpredictable (customers request goods at any time), from
There the importance of disruptive technologies, each fulfills a specific purpose.
Artificial Intelligence is in the center of all of them, being the one that allows the
company to adapt to the change, by analyzing all the information produced in
the production chain and take the most appropriate decisions when knowing the
general operation of the company, And when this determination is reached, it is
necessary that this must be transformed into real actions, for this it is essential
to define the communication network, so nodes have to receive the information,
which machines will execute the action and what effects occur; for which it
is necessary to retransmit the executions of the machines to the general body
control, so that it makes the pertinent corrective actions. The previous process
will be carried out all the time, with the different systems that make up the
company.
Once the Artificial Intelligence analyzes, transmits the orders and receives
feedback, it is able to perform this action with all the devices that are in the
company in real time, which will allow managing the overall operation of the
system, when determining in advance the series of processes that must be exe-
cuted, At the time of any problem of any kind in the system, the solution will be
sought, in case of a direct human intervention: to make a mechanical change of a
device, reprogramming a robot, register a drone, electrical failure, earthquakes,
etc. reprogramming of the master production program could be done to define
new delivery dates, thereby reducing losses.
In the production lines we currently have areas with industrial robots [5]:
90% Welding, 50% Painting, 80% Die-cutting, 20% Assembly; but new robots
such as collaboratives robots are appearing, which can cover the percentage that
is missing, if this happens, the productivity of these areas must necessarily be
increased. In the case of humanoid robots, they could be applied in areas of
customer service at the window, in bureaucratic offices, sales agencies, etc., so
that the procedures can be done faster, not having the need to move to find
information, for being connected wirelessly to the general database.
In the case of 3D printing you have many of opportunities in the printing
of: car bodies, cabinets, casings, metal parts, etc., you must be aware that at
this time the construction time is very long; but it has the advantage that once
the design order is given immediately, the execution order is given, which makes
them ideal for new products, metal parts with hollow interiors, parts that are out
of the market, construction of spare parts in advance, in general manufacturing
of any pieces with irregular shapes that the only way is that this type of device,
which, once they leave the printer can begin with the assembly of the other
parts.
A problem in the manufacturing lines is that to make a product it is necessary
to have a car body that contains all the parts that it will carry, for which
manufacturers use different techniques to build these, such as injection processes
for the car of electronic products, car body building stamping, die-cutting with
oxyacetylene for metal structures, etc., which consume a lot of time for its design,
construction and installation, which makes any mistake or design change very
expensive.
On the other hand, if we use 3D printing, it would allow us to build the entire
car body of the product, make the necessary adjustments, change the design, in
an economical and fast way; if 3D printing goes hand in hand with artificial
intelligence [4], we would obtain additional potential, having the ability to see
what the market demands in real time, changes in production volume, changes
in design, order and cancellations, resulting in a decrease in costs, which will
give the company a competitive advantage.
A very revolutionary application is represented by the digital twins companies
[7], that which consists of a physical part (the product) and the virtual part
(Artificial Intelligence), within the virtual part there is a set of designs that
correspond to the different models that it has The company, at the time the
customer makes the order for any of them, the system gives a code which is
associated with the following data: customer, model and place of delivery; This
production order is transferred to the corresponding plant to define the delivery
time, in turn the plant assigns the start-finish time and the manufacturing line;
When the scheduled date arrives, the body or cabinet is manufactured in a 3D
print, a motherboard with the code corresponding to the personalization of the
product is placed.
This motherboard is part of the embedded system that will be the one that
will control the different production phases, the most important sensors in man-
ufacturing are RFID [8] (which defines the work cells) and GPS geolocation (for
external logistics), To fulfill this purpose, the communication must be wireless,
the ERP control system with Artificial Intelligence, would track the product in
real time, thus eliminating that of working blindly.
When the product leaves the production line [6], the system will define the
route and the means of transport required; The GPS geolocation sensor will keep
you informed of the time and distance traveled, in this way you can check the
total time consumed, as well as those unforeseen for future improvements in the
same process, so that in external logistics blind work is eliminated.
Once the product reaches the customer’s hands, Artificial Intelligence will
be able to monitor [14] the behavior of the product, the habits of use, the
energy consumption, the general performance of the product, for this it will
use the different sensors that have the same such as: temperature, vision, lis-
tening, speaking, vibration, GPS, RFID, etc., so that with this information the
buyer can be indicated about the required maintenance, software updates, avail-
able applications and product life time (which is depending on mechanical use),
so that suitable precautions may be taken. As the manufacturer can sell thou-
sands, hundreds or millions of products, Artificial Intelligence along with Big
Data, Analytics and the cloud, can make a comparison of the use of all its prod-
ucts, the failures, the communication problems presented, the spare parts used,
satisfaction surveys, comments on social networks of the product, sales areas,
characteristics of the population, religion professed, culture and comments on
the company, This way you can make a very approximate profile of the think-
ing of customers, to know what you like or dislike, improvements and possible
models for different ages of the population.
All of the above can not be possible today because of many countries, as well
as companies suffer from 5G communication (24 to 100 GHz transmission range),
which can work with massive information and transmit it in milliseconds; The
most extreme case is the ability to receive information from 100 devices in a
square meter, which is inferred from having a hyperconnected world.
3 Results
Companies are constantly changing due to the emergence of soft and hard tech-
nologies, which cause them competitive disadvantages, these have to be elim-
inated in order to remain in the market, for this purpose they should not be
lost in the multiple technological alternatives that are they present, for which
it is important to identify their strengths and weaknesses, to have the right
human talent and the most efficient technological option to create the competi-
tive advantage.
It is important to keep in mind that artificial intelligence, intelligent
machines, robotics, digital twins and 3D Printing are in the phase of growth
in the curve of the technological life cycle, so it has to start to be implemented
in the productive processes, with the purpose of knowing their capacities and
deficiencies, as well as defining the technological profile of the collaborators, the
training programs, the life plan, as well as the most appropriate organizational
design for the fulfillment of the objectives and goals of the company.
We must also keep in mind that we have other technologies such as, cloud
computing, big data, mobile internet, augmented reality and virtual reality, that
allow us to increase automation and efficiency in production processes, which are
complementary to the previous ones and which together will allow us to solve
complex problems of the everyday world, as would be the case with smart cities.
Which generate a large number of solutions to problems in the fields of: health,
safety, transport, housing, energy, water and food, among others.
The purpose of the 4.0 revolution is to change the current paradigms, of how
we understand technology and its future implications, to have a more inclusive
world, where ethics plays an important role in defining its regulation to maintain
a harmony between humans and machines.
Data analytics and the use of the cloud will be part of a complex and intel-
ligent process that allows for reliable and well-processed data that helps genetic
algorithms work with quality information, generating reliable solutions for the
current requirements of the company.
For the production of a set of different and personalized goods, a real-time
transmission of information is necessary, as well as the control of the process, in
order to comply in a timely manner with what the market demands, and also
consider that the consumer does not follow any rules at the time of establishing
their requirements, this causes that you constantly have to be planning, to adapt
the production to these needs.
It is very important that companies determine the time that consumes the
different custom goods, so that in their computer platforms customers obtain the
delivery time and the price, in a few seconds, which will imply a communication
with the entire value chain (suppliers, manufacturing and distribution), in this
way comply with the agreement.
By having a more informed client, it becomes more demanding, for the reason
that in a few minutes you can visit different Web pages of each of the companies
that interest you, making it necessary to have your own vision that distinguishes
it from the rest of the competition, otherwise the chances of remembering it will
be minimal.
4 Discussion
We are getting to the point that if we want to be competitive it is important
to use smart manufacturing, cloud-based manufacturing, cognitive manufactur-
ing, etc. Which have a very close relationship with artificial intelligence, we are
allowed to design and develop intelligent systems that contribute to the creation
of machines, which have the ability to collect and analyze data, and communicate
with other systems that have the characteristic of solving complex problems.
We are in a social transformation at this time, which is reflected in a constant
change in the way of doing and seeing the world, the new generations of young
people are more qualified for work in smart factories, they are accustomed to
change, so they are more familiar with these new technological innovations, as
well as their use, application and new solutions, so they will continue being a
factor of change.
Many of the ways in which the company works will tend to disappear and
will be replaced by intelligent methods, which will be constantly changing, for
which they must have a life of their own, to deal with unexpected changes in the
production processes, models business and agile project management because it
will demand capacity and speed of response to the demand for new solutions.
With respect to manufacturing companies they have to be workplaces that
allow workers to have a good work culture, which will be reflected in the develop-
ment of their potential, with organizational structures that have as their purpose
innovation, caring for the planet and respect for the different ways of thinking,
all this within the framework of tolerance and respect for others.
We have 5 years to define how manufacturing companies 4.0 will be, in what
they have to do with their organizational structure, this is very important, since
it defines the rights and obligations of all those involved in the manufacture of
products, as well as with the plant distribution, which until now is online, which
implies the sequencing of activities. With new technologies such as: artificial
intelligence, robotics, 3D printing, Big Data, the cloud and intelligent machines,
we must think of other ways of organizing production, technological possibilities
allow.
5 Conclusions
Although much of the current literature indicates that the future of humanity
will be of abundance and mass unemployment, where machines will have control
of all daily activities, it should be remembered that at the time a technologi-
cal development is used in general, the people learn to see its advantages and
disadvantages, if this happens we can say that there is a real understanding of
it, so it will take another technological revolution to solve the new needs of the
population.
We must also bear in mind that at present people only use 10% of our brain
mass, and that as we use it more, it will necessarily have to increase the percent-
age that is not used, so we can be in a future beings that can travel in the galaxy
and see how it is and the machines will continue to be the tools that allow us to
achieve this purpose and generating the right conditions, so that human beings
can inhabit, eat and live on these planets.
The use of disruptive technologies, such as the Artificial Intelligence, is only
paving the ground for the aforementioned to happen and for human beings to
abandon the phase of scarcity of resources, hunger, injustices, etc. for a new one
where the search for happiness, finding our true selves and facing any unknown
challenges ahead will be the ideals of this new era.
Acknowledgments. To the Academic Community of the Master’s Degree in Indus-

trial Engineering of the Postgraduate Studies and Research Section at Professional
Interdisciplinary Unit of Engineering and Social and Administrative Sciences from the
National Polytechnic Institute.
References
1. Aggour, K., Gupta, V., Ruscitto, D.: Artificial intelligence/machine learning in
manufacturing and inspection: a GE perpective. MRS Bull. 44, 545–558 (2019).
Materials research society. www.mrs.org
2. Bartodziej, C.: The Concept Industry 4.0, 1st English edn. Springer, Wiesbaden
(2017). https://doi.org/10.1007/978-3-658-16502-4
3. Brau, S.: Lean 4.0 Manufacturing, 1st English edn. (2016)
4. Bukkapatnam, S., Afrin, K., Dave, D., Kumara, S.: Machine learning and AI for
long-term fault prognosis in complex manufacturing systems. CIRP Ann. - Manuf.
Technol. 68, 459–462 (2019)
5. Ford, M.: Rise of the Robots, 1st English edn. Basic Books, New York (2015)
6. Hu, L., Miau, Y., Wu, G.: iRobot-Factory: an intelligent robot factory based on
cognitive. Future Gener. Comput. Syst. 90, 569–577 (2019)
7. Joyanes, l.: Industria 4.0, 1a edición en español. Alfaomega, México (2018)
8. Knapciková, L., Balog, M.: Industry 4.0: Trends in Management of Intelligent
Manufacturing System, 1st English edn. Springer, Cham (2019). https://doi.org/
10.1007/978-3-030-14011-3
9. Leflar, J.: Practical TPM, 1st English edn. Productivity, Portland
10. López, J.: La gran transición, 1a edición en español. Fondo de cultura económica,
México (2018)
11. Monden, Y.: Toyota Production System. 4th English edn. CRC Press, Boca Raton
(2012)
12. Moreira, L., Li, W., Fitzpatrick, M.: Supervision controller for real-time surface
quality assurance in CNC machining using artificial intelligence. Comput. Ind. Eng.
127, 158–168 (2019)
13. Ostrosi, E., Fougeres, A.: Intelligent virtual manufacturing cell formation in cloud-
based design and manufacturing. Eng. Appl. Artif. Intell. 76, 80–95 (2018)
14. Qin, J., Chian, L.: Advances and opportunities in machine learning for process
data analytics. Comput. Chem. Eng. 126, 465–473 (2019)
15. Ustundag, A., Cevikcan, E.: Industry 4.0: Managing the Digital Transformation, 1st
English edn. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-57870-5
Software Engineering and Education
Mobile Application for the Support
in the Learning of the Alphabet, Verbs
and Pronouns of the Mexican Sign
Language Based on Augmented Reality
A. Gordillo-Ramı́rez1 , O. Alonso-Cuevas1 , D. Ortega-Pacheco2(B) ,

and U. Vélez-Saldaña1
1
National Polytechnic Institute, ESCOM,
Av. Juan de Dios Bátiz s/n esq. Av. Miguel Othón de Mendizabal, Col. Lindavista,
Gustavo A. Madero. C. P., 07738 Mexico City, Mexico
almagoram@gmail.com,osvaldo.alocue@gmail.com,ulises.velez@gmail.com
2
National Polytechnic Institute, UPIICSA,
Av. Té #950 esquina Resina, Col. Granjas, Iztacalco, C.P.,
jdortegap@ipn.mx
Abstract. In this paper we present a mobile application for android

operating system focused for children’s, for the support in the learning of
the alphabet, verbs and pronouns in the Mexican Sign Language context.
The parents or teacher will be able to download the markers for each
Lesson or Game. Once the markers are available, the child will be able
to carry out his activity with the support of the augmented reality.
Keywords: Augmented reality · Mexican Sign Language · Mobile

learning · Mobile application
1 Introduction
An alternative to address the child learning process for Mexican Sign Language
is through the technology we currently have, since the fast advancements of
technological developments allows to find wide possibilities in the learning envi-
ronment, such as the use of mobile devices and augmented reality. There are ben-
efits of using mobile devices in the learning process, such as: improves student’s
autonomy, facilitates feedback and helps to understand topics with multimedia
elements (by augmented reality, for example).
This learning way is good for topics like: Problems solving, to acquire skills,
languages, continuous learning, exploratory and transversal competences, games,
science and model representation. This learning method is called Mobile Learn-
ing and it is a new strategy for education. In this learning method, the materials
are important: videos, conceptual maps, images, models, conferences, simula-
tions, test, etc. The contents should be optimal and suitable for mobile devices.
https://doi.org/10.1007/978-3-030-33229-7_16
184 A. Gordillo-Ramı́rez et al.
In M-Learning the automated organization of content (topics) is promoted more

in a similar manner when the teachers work with learning objectives, and this
strategy recommends sectioning the topics in little units, it forms complete infor-
mation, learning activities and contextualization elements [1].
In the educational context, there are previous works that have endeavored
to support learning as in [2], which describes how augmented reality can be
used to teach concepts in the field of health, using a mobile device; in [3] shows
how augmented reality is used to support children with learning disabilities.
In the specific context of sign language learning, there is, for example, in [4],
where augmented reality is used on a PDA for the teaching of Brazilian Sign
Language, focusing only on the alphabet; in [5] the elements of augmented reality
are used through a personal computer or a tablet, using an exercise book to
support learning, which contains the markers that trigger augmented reality to
visualize learning objects, in this case the alphabet, and in [6] the elements for
the augmented reality are words and letters of the Arabic Alphabet in an mobile
environment. With the above we can realize that in terms of learning, in regard
to Sign Language, the works focus on the alphabet and not on more language
elements such as verbs, pronouns, articles or phrases, which are important for
learning that language, independently of the country.
The present work described below, pretend to be a mobile application tool,
that intends to integrate basic elements for child learning of Mexican Sign Lan-
guage such as: Alphabet, Pronouns and Verbs, based on augmented reality and
its use in mobile devices with Android operating system.
2 Learning Environment
The learning environment is based on the use of a mobile application, whose
development focuses on the concepts of Mobile Learning, Gamification and Aug-
mented Reality, as described in Fig. 1. The application is designed for children
in first year of primary school, and pretend to be a supporting tool for the learn-
ing of the alphabet, verbs and pronouns, that are the base for more complex
structures in the Mexican Sign Language context.
Each child must have access to a mobile device with android operating sys-
tem, which according to the statistics obtained from [7], for May 2019 has 82.28 %
of use in mobile devices in Mexico.
Each learning element (alphabet, verbs or pronouns) has a lesson, which is
carried out by the child and supervised by the teacher in the classroom or by
one of his parents in other place. Each lesson is based on the use of augmented
reality technology through printed markers. Once the child completes a lesson,
the parent or teacher can record the evaluation corresponding to the child’s
performance. There is also a game for the alphabet, which is not intended to
make an evaluation, the aim is to motivate the child to practice the elements of
the corresponding lesson of alphabet. The approach for the mobile application
development, are described as follow.
Mobile Application for the Support in the Learning of the Alphabet 185
Fig. 1. Learning environment
3 Mobile Application Development

We have use the agile development approach: Mobile-D (more detail in [9]), witch
consists in five stages: Explore, Initialize, Productionize, Stabilize and System
test and fix. We developed the mobile application as follows:
Fig. 2. Mobile application architecture
– Explore. In this stage, we define the software scope (requirements), user

environment and the architectural design (See Fig. 2).
– Initialize. In this stage we prepare and verify all critical development issues
for requirements implementation and carried out the concept proofs for aug-
mented reality, 3D models and android application development.
– Productionize. In this stage we implement the functionality associated in
each module for the architecture design, applying an iterative and incremental
development cycle.
– Stabilize. In this stage we ensure the quality of each implementation, and if
it was required, we carried out the necessaries adjustment.
– System test and fix. In this stage, we obtained the team feedback for
the application functionality and fix the found defects for the stabilization
between modules. Finally we carried out the users acceptance tests for the
mobile application, in the learning environment.
4 Access Control
Before a child start the mobile application on a device, one of the parents or
the teacher must be registered, and then log in and activate the corresponding
activity, either a lesson and its subsequent evaluation, or a game. The account
Information and Log in, is showed in Fig. 3.
Fig. 3. Access control: Account Information and Log in
5 Augmented Reality
The augmented reality in the mobile application is based on the use of the
ARToolKit API, in his 5.3.3 version, for more detail see [8]. The general operation
of this module is showed in Fig. 4.
Each element is described as follow:
– Marker. Has the design showed in Fig. 5, that contains the dimensions and
sections for each marker, and represents the relationship between signified
and signifier, in the Mexican Sign Language context. In the example, the bee
image (Abeja in Spanish) is the signified, and the image with the letter A and
the corresponding Mexican Sign Language representation, are the signifier.
This concepts presents in the marker, are crucial for language acquisition
process.
Fig. 4. Augmented reality operation
Fig. 5. Marker design and example for letter A
Fig. 6. Feature map image: edge points in green crosses and textured areas in red
squares (Color figure online)
– Marker identification. The camera in the mobile device, is the responsible

for processing the marker. For each image frame, the object tracking method
(part of the Natural Feature Tracking), is used to for process natural inter-
est points of an image such as: texture-patches, edges, color blobs, etcetera.
This method generate a image with a feature map with this interest points
(see Fig. 6). The input marker identification, is performed by the comparison
between his feature map image processed in each image frame by the camera,
with all feature map images stored in the mobile application, referenced by
the markers.dat file, and corresponding to all recognizable markers.
– Search 3D Model. If the input marker is identified, the module should
search the corresponding 3D Model, that is referenced by the objects.dat file.
This model is the 3D graphical representation of the Mexican Sign Language
signifier corresponding in the identified input marker.
– 3D Model on Marker. In this stage, the module identifies the input marker
ubication, and show on the mobile device screen, the 3D Model over the input
marker. At this point, the user can move the input marker to left or right,
and the 3D Model reacts to this movements showing the corresponding view.
To accelerate the augmented reality processing, OpenSceneGraph (OSG).
6 Lessons
The lessons corresponding to the alphabet, pronouns and verbs in the application
are described below, in the order that which they are executed:
1. Alphabet. We use the Mexican Spanish alphabet: 28 markers, and we con-

sidered vowels, consonants and an example of a double letter, however we
show the vowels without spelling accent and without umlauts. Each marker
has the signified with an image that indicates an object, it also has the first
letter of the object that kids can view in the image. In the 3D models for
this lesson, we use only the hand because for the alphabet is not necessary
to show the complete body, so the model must have a hand with five fingers
and the corresponding arm that allows the movement of the Sign. The child
performance is evaluated in the mobile application by the parent or teacher.
An example is showed in Fig. 7.
Fig. 7. Example of alphabet lesson and evaluation

2. Pronouns. The second lesson is personal pronouns in Mexican Spanish Lan-

guage, and we have 9 pronouns (if we think about feminine and masculine
pronouns) therefore in LSM the signs for female and male are different each
other. We decided to have only 5 signs to put on view, and each marker has
signified so in this lesson we did 3 people to represent the pronouns and has
signifier represented by the word and sign. In the special case of the pronoun
“Nosotros”, there are three ways to do the same sign “Nosotros”, but the sign
that we show is the most actual and it is used by young people. The child
performance is evaluated in the mobile application by the parent or teacher.
An example is showed in Fig. 8.
Fig. 8. Example of pronouns lesson and evaluation
3. Verbs. The last lesson is verbs, we have knowledge of Spanish Language has
a lot of verbs, but first we were taking into account that in Spanish, the
LSM verbs do not indicate mode or time, also in LSM, the verb “Ser o estar”
(To be) is very rarely used, for these reasons we selected some verbs related
to the school and then we carefully chosen some of them that have easier
movements. When we selected the five verbs, we designed the markers only
with signifier represented by the word and sign because the marker is too
small to include much information. In this time, we have be careful with the
position and structure of the hands because some verbs can be confused with
other sign. The child performance is evaluated in the mobile application by
the parent or teacher. An example is showed in Fig. 9.
Fig. 9. Example of verbs lesson
7 Game
This game consists on asking for 20 random letters to child, each letter at time,
so they have to place the correspondent marker in front of the camera, allowing
the recognition of the marker and verifying if it match with the asked letter,
therefore the child can interact with the markers and the mobile application.
To implement the game mechanics we had to add a list containing the letters
and it’s corresponding marker index, the application randomly extracts a letter
and asks for it in the screen, also, a 3D model of a picture frame is displayed
to provide feedback, so the user knows the application is detecting a marker,
besides, we added a GIF image of the letter’s sign as assistance to recognize
the correspondent marker. We implemented a reward system based on trophies
represented by candies, if the child place the 20 letters correctly in a time less or
equal to two minutes, three candies will be provided, if the letters are placed in
a time higher than 2 min and less or equal to 3 min, the child will be rewarded
with two candies, and finally, if the child place the letters in a time higher than
3 min, the reward will be one candy. An example is showed in Fig. 10.
Fig. 10. Alphabet game

8 Discussion and Future Work

With this application, it is shown how augmented reality can support more
actively on the learning of elements of Mexican Sign Language, integrating ele-
ments that are the basis of it: Alphabet, Pronouns and Verbs; this leaves the
basis for later reaching a mobile application that helps to represent more com-
plex phrases, where the execution of a sign has a dependence on the position
of the body where it is performed. An additional element will be to be able
to use pattern recognition techniques to be able to automatically evaluate the
performance of a child in the use of the mobile application.
Acknowledgements. The authors would like to thank to the National Polytechnic

Institute (IPN) of Mexico, for the support in the development for this project.
References
1. Ramı́rez Montoya, M.: Dispositivos de mobile learning para ambientes virtuales:
implicaciones en el diseño y la enseñanza. Educación de la Universidad Virtual del
Tecnológico de Monterrey. Monterrey, México (2008)
2. Aurélio Galvão, M., Roberto Zorzal, E.: Augmented Reality Applied to Health
Education. In: 2013 XV Symposium on Virtual and Augmented Reality. IEEE
(2013). https://doi.org/10.1109/SVR.2013.54
3. Vinumol, K.P., Chowdhury, A., Kambam, R., Muralidharan, V.: Augmented reality
based interactive text book: an assistive technology for students with learning dis-
ability. In: 2013 XV Symposium on Virtual and Augmented Reality. IEEE (2013).
https://doi.org/10.1109/SVR.2013.26
4. Rabelo Nazareth, D., Dos Santos Alencar, M., José Francisco de Magalhães Netto:
ELRA - teaching brazilian sign language using augmented reality. In: 2014 XVI
Symposium on Virtual and Augmented Reality. IEEE (2014). https://doi.org/10.
1109/SVR.2014.37
5. Cadeñanes Garnica, J.J., Arrieta, M.A.G.: Augmented reality sign language teach-
ing model for deaf children. In: Omatu, S., Bersini, H., Corchado, J.M., Rodrı́guez,
S., Pawlewski, P., Bucciarelli, E. (eds.) Distributed Computing and Artificial Intel-
ligence, 11th International Conference. AISC, vol. 290, pp. 351–358. Springer,
Cham (2014). https://doi.org/10.1007/978-3-319-07593-8 41
6. Al-Megren, S., Almutairi, A.: Assessing the effectiveness of an augmented reality
application for the literacy development of arabic children with hearing impair-
ments. In: Rau, P.-L.P. (ed.) CCD 2018. LNCS, vol. 10912, pp. 3–18. Springer,
Cham (2018). https://doi.org/10.1007/978-3-319-92252-2 1
7. Mobile Operating System Market Share in Mexico - May 2019 Homepage. Stat-
Counter. http://gs.statcounter.com/os-market-share/mobile/mexico. May 2019
8. ArToolkit Homepage. http://www.hitl.washington.edu/artoolkit/. May 2019
9. Mobile D. Electronics Agile Homepage. http://virtual.vtt.fi/virtual/agile/mobiled.
html (2019)
10. Neumann, U., You, S.: Natural feature tracking for augmented reality. In: IEEE
Transactions on Multimedia, vol. 1, No. 1, March 1999. https://doi.org/10.1109/
6046.748171
Augmented Reality with Swift in ARkit
and Their Applications to Teach Geometry
Eduardo Eloy Loza Pacheco1(&) , Mayra Lorena Díaz Sosa1(&) ,

and Miguel Jesús Torres Ruiz2
1
Universidad Nacional Autónoma de México, Acatlán Edomex,
{eduardo.loza,mlds}@apolo.acatlan.unam.mx
2
Instituto Politecnico Nacional, CDMX, Mexico City, Mexico
mtorres@ipn.mx
Abstract. Augmented Reality technologies are now entering in a maturing

process. We can see major computer technologies companies developing it. For
example Google with ARCore born in 2018 or Apple with ARKit born in 2017.
These options allow us to develop a very friendly augmented reality environ-
ment with a few steps and a very friendly computer-human interaction. An iOS
App needs to be developed using Swift a young language born in 2014. This
work shows the application of ARkit to ease the learning of Geometry in R3
space.
Keywords: Augmented reality Teaching geometry Swift and ARkit
1 Introduction
In the past two decades, we can see several technologies of virtual reality and aug-
mented reality such as OpenGL, VRML [1], X3D that is web technology [2]. These
technologies have a limited impact on educations because of the availability of com-
puter equipment. Twenty years ago the reduction cost of computer technologies per-
mits the democratization of access to the Internet. The reality now is different mobile
devices are part of the fourth revolution [3]. Two technological firms Apple and Google
are developing two different and independent implementations called ArKit and
ArCore respectively to exploit the augmented reality market, which is becoming
standards [4]. Education is one of the greatest beneficiaries thanks several students have
access to mobile devices. It is possible to teach several mathematics topics with the
help of augmented reality. Specifically, geometry that is one of the main topics in
engineering at Universities because develops our abilities of spatial reasoning.
Examples of applications we can find in GI Science where we need to teach Cartesians
to polar transformations, robotic planning the student will need concepts such as vector,
planes in order to describe the states of an agent, etc. Swift is a programming language
born in 2014 and open to the Open Source community in 2015. The ArKit API was
released in 2017 [5] and multidevice integration in the fall of 2018 [6]. On WWDC
2019 Apple introduced new advances on Augmented Reality, ARKIT 3 and Reality

https://doi.org/10.1007/978-3-030-33229-7_17
Augmented Reality with Swift in ARkit and Their Applications 193
Composer. The present very interesting novelties such as: People Occlusion, Motion
Capture Multi-Face Tracking, Simultaneous front and Back camera, Collaborative
Sessions, y and improvement on image recognition. People Occlusion allows to insert
animation in front of people or hide some part of their bodies. Motion Captures scan
the body of a person to track their movement and generate a 3D model. Front a Back
camera are now available to captures images in order developers can program more
interactive applications. Face tracking allows to sense the face of a person so we can
develop more interesting applications, these features is presents in all devices with iOS
13 or higher [7]. Reality composer is a full framework designed to integrate all
applications not just. For iOS but for MACOS, integrates Animations and audios, can
record and play [8] (Fig. 1).
Fig. 1. Arkit 3, motion capture taken from: [9]
2 Didactic Strategy
Virtual reality and augmented reality allow us to visualize elements and environments
as never seen before. Although education has been enriched by these tools, there are
few documented cases of its use to promote the learning of abstract concepts of
mathematics and exact sciences. But these environments can be useful to create
learning objects for analytic geometry under a playful approach. In Fig. 2 we can see a
Virtual environment develop in Java 3D, that we have developed in class.
194 E. E. Loza Pacheco et al.
Fig. 2. Virtual reality Tic-Tac Toe develop in Java 3D
In contrast a we can see, the same game in Augmented reality using Arkit in Swift
in Fig. 3. The environment can be combined with elements in reality that teacher can
be used to add additional information, with the help of another element such as
blackboard or projector.
Fig. 3. Augmented reality Tic-Tac Toe developed ARKit
3 ArKit Implementation
Arkit implementation is simple and reduces the complexity that can be found in many
programming languages such as Java using the API Java 3D [10], or VRML [1]. For
example in Java, the programmer has to write large sums of code, in VRML the code
has no expressivity. Instead Swift keeps important concepts such as the definition of a
root object and all objects heritage of it. As the majorities of all Graphics programming
languages, As we can see in Fig. 2 shows the design of the virtual world is the same as
in VRML. Where the “transform” node has the regular transformation actions such:
translation, rotation and scale and geometric attributes. Similar to VRML in Java 3D
we can see the same structure (see Fig. 4).
Fig. 4. (a) Semantic net of a scene for VRML and (b) Java 3D structure.
However Swift can create scenes from programmatically as we can see in the
previous images. Or Visually by adding objects in art.scrassets View. Here we can
modify objects such: color, transparency, transformation (translation, scale, rotation).
As we can see in Fig. 5.
Fig. 5. Sphere transformation in Swift
The programmatical we can see ARKit divides the properties of the graphic object
in material, shape, and appearance in one leaf of the root object of the scene and the
transformation properties in another like scale, translation, and rotation. Also, we can
be found the same object definitions of geometry objects such, textures, transparency
and illumination. In Fig. 6(a) we can see a fragment of the code of a part of a scene in
Java 3D versus the Swift implementation.
Fig. 6. (a) Java code in Java 3D and (b) Swift code in Swift ARKit.
Using ArKit in Geometry. The API allows us to create an AR scene and combine
them with real-world images. Additionally, as is common in AR implementations, AR
Kit has an implementation that permits image recognition. For this purpose, we create a
project in XCode (New iOS Project/Augmented Reality App). So we can see a envi-
ronment like this un Fig. 7. Then we add a new AR resource Group, so we can add the
set of images we would like to recognize (New AR Resource Group), the result is
shown in Fig. 8. The images are the place in the real world where Swift will place the
AR scene [11, 12].
Fig. 7. Xcode environment using ARkit
Fig. 8. AR Resource groups.
After that we need to enable the application the possibility to recognize images, as
is seen in line 43 of Fig. 9 in the function viewwillAppear(…), we will need to create
the object that can discover the images. Also, we need to specify the Folder name
where the images are stored (The image should have good resolution). Then we just
need to add different images and program a different behavior for everyone.
Fig. 9. Enabling tracking configuration object in line 43.
Having built the Environment, we need to select the size of our screen in order to
develop for that size. In Fig. 10(a), we can see the different size XCode allow us to use.
In Fig. 10(b) We can select the compatible technology we would use.
Finally, with these few lines of code, it is possible to develop an application that
works with hybrid objects. Where part of the implementation is in virtual scenes of AR
and others in a physic document. This gave us the opportunity to integrate mobile
devices and conventional educational resources. For example, whether we need to
explain planes and surfaces of Geometry. Part of the learning process can have a
theoretical background, and the practice can be the human-machine interaction with the
virtual world. In Fig. 11 we can see a small learning object applied to this matter.
Fig. 10. (a) Different iOS devices (b) Different version of devices.
Fig. 11. AR scene with a physic document.
4 Conclusion
Modern technologies of AR are becoming very popular despite the fact there are new.
In the case of ARKit of Apple, has been eased the form we develop AR Applications.
Also, with the decreasing cost of mobile technologies. Today we have a great
opportunity to integrate AR to a variety of applications in areas where the society can
be benefited. The areas of applications are wide such as education, GIS, robotic
planning, engineering, spatial reasoning, etc.
Acknowledgements. The authors would like to thank to CONACyT and DGAPA-UNAM

(PAPIME PE 304717) for the funds to support this work.
References
1. Carey, R., Bell, G.: The Annotated VRML 2.0 Reference Manual, vol. 15. Addison-Wesley,
Reading (1997)
2. Web3d.org Homepage. What is X3D?| Web3D Consortium. http://www.web3d.org/x3d/.
Accessed June 2018
3. Toffler, A., Alvin, T.: The Third Wave, vol. 484 (1980)
4. Stallings, W.: Network and Internetwork Security: Principles and Practice, vol. 1. Prentice
Hall, New Jersey (1995)
5. Apple Inc. Homepage, ARKit. https://developer.apple.com/arkit/. Accessed June 2019
6. Apple Inc. Homepage, ARKit. https://developer.apple.com/documentation/arkit/creating_a_

multiuser_ar_experience. Accessed June 2019
7. Apple Inc Homepage., ARKit. https://developer.apple.com/augmented-reality/arkit/.
Accessed 2 July 2019
8. Apple Inc., ARKit. https://developer.apple.com/augmented-reality/reality-composer/.
Accessed July 2019
9. Andy O`Sullivan video page. https://www.youtube.com/watch?v=VKSFVLO_qaI. Acces-
sed June 2019
10. Pratdepadua, J.: Programación en 3d con java 3d, Ra-Ma (2003)
11. Apple Inc. Homepage, Intro to App development with Swift, Apple Education. Accessed
2018
12. Apple Inc. Homepage, App development with Swift, Apple Education. Accessed 2018
Computer Tool Prototype for the Selections
of Views to Materialize in Data Cubes
and Frequent Pattern
Elizabeth Moreno Galván(&) and Enrique Alfonso Carmona García
UPIITA-IPN, Av. Instituto Politécnico Nacional 258,

Gustavo A. Madero, 07340 Mexico City, Mexico
elizabeth.moreno.galvan.05@gmail.com,
eacarmona860920@gmail.com
Abstract. The widespread availability of new computational methods and tools

for data analysis represents a great challenge for new students who need to now
the basics. It is difficult the selection of the most appropriate strategy and the
collection of methods in data mining it’s a good example. Some of these
methods require guidelines that may help practitioners in the appropriate
selection of data mining tools. In the present paper it is showed the proposal of a
tool which shows visually the implementation of an algorithm for the selection
of the optimal set of views to materialize from a multidimensional cube, being
the main parameter the materialization cost of those views. The tool allows
defining operations with cubes on different dimensions from a list of business
questions that can be performance on them, all this in order to ease students the
learning and understanding of data cube generation in data mining discipline.
Keywords: Data mining Data cube Educational data mining Big data
1 Introduction
Over the last years, the term data mining has been increasingly used in scientific
literature and implies the use of different methods and tools to analyze large amount of
data. Also, data mining term refers to data sets that are so massive or complex in nature
so that traditional data processing methods and methodologies are inadequate to
effectively analyze them [1, 2]. Data mining and big data has been successfully applied
to different fields of human endeavor, including environmental and pollutant studies
[3–6], disease detection and m-health [7–10], IoT and smart cities [11–13], but it does
not only apply to these fields; these are just some of the recent fields of application of
these techniques, they also apply to economic studies such as price prediction, mar-
keting, or banking operations to name a few.
The data warehouse were created in response to the need to gather all the infor-
mation of an organization in a single database, for analysts and administrators in order
to find valuable data or verify situations that are related to different types of decisions
where it is also necessary to reduce the time to obtain it [14–16]. The generation of a
warehouse requires applying processes of extraction, integration, transformation and

https://doi.org/10.1007/978-3-030-33229-7_18
204 E. Moreno Galván and E. A. Carmona García
cleaning data. The design and construction of a data warehouse is a complex and
delicate task [16–18], since it must be outlined the multidimensional model to be used,
as well as defining the optimal set of views to materialize. Finally, the last process
consists of determining the necessary tools for data visualization.
This paper focuses on the data organization process from a warehouse using a
software tool that automatically designs a multidimensional model and carries out the
materialization of the hypercube to subsequently apply the algorithm of selecting the
optimal set of views to materialize (Harinarayan method) [19]. The tool is also able to
determine the frequent patterns present in the data taking into account a certain support
and confidence, calculating also the association rules. The results of execution of the
software tool developed are analyzed and finally the conclusions are presented.
2 Literature Review
The use of data mining in educational fields is knowing as Educational Data Mining,
which emerges as a paradigm oriented to design models, methods and algorithms for
exploring data from educational settings [20]. In this section are analyzed some fre-
quently used educational data mining tools and techniques.
In [21] a review of works was presented; identifying statistics-visualization and
web mining as a couple of techniques to classify. Related to visualization, four works
are referenced and web mining is split into three tasks:
1. Clustering, classification and outlier detection
2. Association rules and sequential pattern
3. Text mining
Some features that are commonly present in friendly educational data mining tools
for non-technical users, are the standardization of data mining methods and data, the
integration of data mining functionalities and the design of techniques devoted to
educational data mining.
In [20] an interpretation of educational data mining review is present, it shows a
detailed analysis about:
1. Disciplines involved in data mining such as probability, machine learning, statistics,
soft computing, artificial intelligence and natural language, where probability,
machine learning and statistic are the principal approach used.
2. Data mining models, where two models are analyzed: descriptive models (unsu-
pervised learning functions to produce patterns) and predictive models (supervised
learning functions to estimate unknown or future values).
3. Data mining tasks such as clustering, association rules, correlation analysis, produce
descriptive models, classification, regression and categorization generate predictive
models. The most typical tasks are classification and clustering.
4. Data mining methods and techniques such as Bayes theorem, decision tree,
instance-based learning and hidden Markov model are the most popular methods.
Logistic, linear regression, frequencies and hierarchical clustering are the most
popular techniques.
Computer Tool Prototype for the Selections of Views 205
5. Algorithms, equations and frames used for data mining are implemented to mine the
source data. The most popular are K-means, expectation maximization and Naïve-
Bayes are the most used algorithms. Statistical equations are the most used equa-
tions. Several version of Bayesian networks are the most popular frames.
In [22] some of the tools for research and practice in educational data mining are
analyzed, discussing where relevant tools are used in data science communities, but
tools for creating structural equations model and multilevel models are not covered.
The tools analyzed are resumed as follows:
6. Two tools for manipulation, cleaning and formatting the data are discuss (Microsoft
Excel and EDM Workbench).
7. Six tools for determining the test can be conducted, what models can be con-
structed, what relationships can be mapped and explorer and how ca be validating
the findings are analyzed (RapidMiner, Weka, KEEL, KNIME, Orange and SPSS).
8. Tree visualizations tools was analyzed (Tableau, d3js and InfoVis).
In all previous work nothing is mentioned about a tool which shows visually the
implementation of an algorithm for the selection of the optimal set of views to mate-
rialize from a multidimensional cube, neither it is not taken in account the views
materialization cost. Our proposal is a tool that allows defining operations with cubes on
different dimensions from a list of business questions that can be performed on them.
3 Materials and Methods
The algorithm was implemented with the use of programming structures known as a
linked list, which is filled from a data source file and emptied into various views within
a database, forming what is known as the data cube. In a complementary way and to be
able to adequately construct the structure of the data cube a modified B tree is
implemented since each node has several children, it is necessary to take account that a
node, in turn, can also have several parents. In the application interface named the file
selection screen from which the user has to introduce the source file name in CSV
format. If the user does not know the path or source file name, it can be selected from a
file chooser, and once the source file is selected, the application can read the data
source chosen in CSV format.
The data cube is used to represent the facts; in this case, the learning and under-
standing of data cube generation in data mining discipline. A data cube generally is a
multi-dimensional concept; it can be from one to n dimensions. In this study, each
dimension represents a separate measure, whereas the cells in the cube represent the
facts of interest (how to generate the cube). Sometimes cubes hold a few values with
the rest being empty, e.g., undefined, sometimes most or all cube coordinates hold a cell
value of this type.
The tool was developed using Java programming language and was designed to
read any text file sequentially and interpret it appropriately if it contains the data in a
format where the column headings are in the first row separated by commas, as well as
each data within tuple. The program is responsible for calculating the views on the
MySQL database and manage a cached memory file in the form of a simple linked list
and a modified B tree, where each node of the tree constitutes a cuboid or view,
generating both a logical and a visual cube of data, where the calculation of data is
carried out the cardinality of each dimension and its using an algorithm [19, 23] to
determine the views to materialize. The first tests of the execution were made on a data
cube created in a synthetic way; that is, the data was created by a computer program
and the measurements were generated with random values.
4 Analysis and Design of the Multidimensional Model
Once the necessary data have been loaded, we work on the obtained database, so the
schema of a cube is composed of Dimensions (D1, D2, … , DN) that corresponds to the
attributes of the database, obtained from the column headings reading of the file that
contains the data store. The cube generated (Grill L = D1 D2 … DN) contains
as nodes the cartesian product of dimensions D1, D2, … , DN, where each node is
called cuboid and represents potentially useful aggregations [19, 23].
Once the structure is generated, in order to answer any business questions, we must
take advantage of the order in the structure of lattice that store the disk views in its
nodes, besides it may a good option to materialize or pre-calculate all the cuboids,
however the limitation of the size on disc it important to know. On the other hand, it is
recommended to materialize the total base (maximum detail or apex) since with it can
be answer any question and then move to the views with less cost, resulting into less
time in obtaining the answer.
The formulation of business questions can be carried out through expressions of the
structure query language (SQL), based on data obtained from the reading of the data file.
The results of answering expressions or queries are known as view. In addition, oper-
ators can be created in order to work with views or data cubes. So, in a query as follows:
SELECT field; Op ðfieldÞ FROM table GROUP BY field
Where:
– Field: is a subset of attributes within the database called dimensions of the form
D = {d1, d2, … , dn}
– Op(field): is an operation in a numeric type dimension, such as COUNT, SUM,
MAX, MIN.
The views are all those business questions that make use of the GROUP BY clause
of at least one field and can be executed using the apex view. The cost of a view C(v) is
the cost of using that view to evaluated the average question and is proportional to the
size of the view v. So, the construction of the lattice is done as follows:
A ! B si A B
• Goes an arc from A to B, if B is a proper subset of A

• If mean that a question to B (if B does not exist) can be answered by looking only at A
If A \ B 6¼ 0 but A 6 B and also B 6 A; then there is no a arc from A to B
It has been implemented the Greedy algorithm [19] in order to get an efficient views
generation, the algorithm works as it follows: Let C(v) be the cost of a view v, which is
the cost of using that view to evaluate a business question. Due to the disk size, there is
a limit of k views (plus the total or top) to use. After selecting some set S of views, it
has to be defined the benefit of the view v relative to S, denoted by B (v, S). The only
views that benefit from having v materialized are those that can be calculated from v,
including v itself, we call each of them w. So the total benefit is the sum of all the
benefits of the w views. Figure 1 shows the resulting pseudocode.
Fig. 1. The greedy algorithm [19].
Figure 2 shows the Compute cube implementation with average by top-k H-tree-
based implemented for the efficient views generation.
Fig. 2. Compute cube with average by top-k H-tree-based [23].

Where:
• Let C(v) be the cost of sight v
• Due the size of the disc, there is a limit of k views to use
• After selecting some set S of views (which of course includes the apex), the benefit
of the view v relative to S is defined, denoted by B(v, S)
• Calculate B(v, S), the benefit of materializing the view v when the views of the set S
have already materialized
• The only views that benefit from having v materialized are those that can be
calculated from v, including v itself. Its denotated by w
• For each w v, define Bw
• The view w can be calculated from v
• The lowest cost view is selected and denoted as u
• The benefit of Bw is C(v) < C(u), Bw = C(u) – C(v); otherwise Bw = 0
• If v has lowest
P cost than u, the benefit is the difference
• Bðv; SÞ ¼ Bw .
wv
In plain terms, the benefit of materializing v is calculated by looking at how the cost
of evaluating other views, including itself, can be improved.
5 Results
The results obtained in the tool developed when loading the generated synthetic data
are presented below. In Fig. 3 the result of loading the data in the application is
observed.
Fig. 3. Data warehouse reading. Source: Self-made.

Subsequently, the views that can be generated from data cube in which the names
of the dimensions are replaced by an alias is represented in Fig. 4.
Fig. 4. Generation of data cube and its possible views. Source: Self-made.
Having all possible views to generate, we proceed to make the analysis of those that
are suitable to materialize through the use of the proposed algorithms. Each view
represents a set of fixed dimensional data in relation to a variable dimensional data. The
frequent patterns are highlighted in the interface and determined by stablishing the
required values of minimum support as Fig. 5 shows.
Fig. 5. Frequents patrons finding. Source: Self-made.

Finally, the existing association rules are determined by stablishing the required
values of minimum confidence as Fig. 6 shows.
Fig. 6. Associations rules finding with min support = 3 and min conf = 80%. Source: Self-
made.
The data growth is a factor that is always present in any data warehouse and the
analysis of data cubes is an area that is still being studied, and there are many types of
algorithms that address the problem in several ways. However, although the multidi-
mensional design of data benefits the way in which the data cube is materialized, the
null values within the measurement data, provide a greater benefit and a much higher
performance before the queries.
In this article, different test is presented, with combinations of four and eight
dimensions with different degrees of cardinality and associated measure. The results
were presented with respect to the processing time. As a general conclusion, it can be
inferred that the best scenario occurs when the dimension with the lowest cardinality is
in the outermost area of the data cube and the dimension with the highest cardinality is
in the innermost area. Visual learning has shown great benefits in teaching environ-
ments, the tool being visual provides the student with the ability to abstract and
organize information, allowing him to relate the complexity of data analysis in a natural
way and faster that conceptually as has been done traditionally.
Some works that are considered interesting are:

• Accelerate the load of the base to the structure by means of chunks if there is limited
memory.
• Improve the capture interface and the result interface.
• The capture of the expression that defines the business questions is not simple for a
business analyst and the answer is delivered in text format, that makes difficult the
interpretation.
• To include some other analysis algorithms implementations in order to show their
performance and the comparison between them.
References
1. Bellazzi, R., Zupan, B.: Predictive data mining in clinical medicine: current issues and
guidelines. Int. J. Med. Inform. 77(2), 81–97 (2008). https://doi.org/10.1007/978-3-540-
30576-7183
2. Wong, Z.S.Y., Zhou, J., Zhang, Q.: Artificial intelligence for infectious disease big data
analytics. Infect. Dis. Heal. 24(1), 44–48 (2019). https://doi.org/10.1016/j.idh.2018.10.002
3. Wang, J., Boesch, R., Li, Q.X.: A case study of air quality - pesticides and odorous
phytochemicals on Kauai, Hawaii, USA. Chemosphere 189, 143–152 (2017). https://doi.org/
10.1016/j.chemosphere.2017.09.045
4. Wang, Q., Wang, J., He, M.Z., Kinney, P.L., Li, T.: A county-level estimate of PM2.5
related chronic mortality risk in China based on multi-model exposure data. Environ. Int.
110, 105–112 (2018). https://doi.org/10.1016/j.envint.2017.10.015
5. Uni, D., Katra, I.: Airborne dust absorption by semi-arid forests reduces PM pollution in
nearby urban environments. Sci. Total Environ. 598, 984–992 (2017). https://doi.org/10.
1016/j.scitotenv.2017.04.162
6. Bari, M.A., Kindzierski, W.B.: Ambient fine particulate matter (PM2.5) in Canadian oil
sands communities: levels, sources and potential human health risk. Sci. Total Environ. 595,
828–838 (2017). https://doi.org/10.1016/j.scitotenv.2017.04.023
7. Vijayakumar, P., Ganesh, S.M., Deborah, L.J., Rawal, B.S.: A new SmartSMS protocol for
secure SMS communication in m-health environment. Comput. Electr. Eng. 65, 265–281
(2018). https://doi.org/10.1016/j.compeleceng.2016.11.016
8. Kazemi, Y., Mirroshandel, S.A.: A novel method for predicting kidney stone type using
ensemble learning. Artif. Intell. Med. 84, 117–126 (2018). https://doi.org/10.1155/2018/
6798042
9. Echeverría, M., Jimenez-Molina, A., Ríos, S.A.: A semantic framework for continuous u-
health services provisioning. Proc. Comput. Sci. 60(1), 603–612 (2015). https://doi.org/10.
1016/j.procs.2015.08.187
10. Acharya, U.R., et al.: Data mining framework for breast lesion classification in shear wave
ultrasound: a hybrid feature paradigm. Biomed. Sig. Process. Control 33, 400–410 (2017).
https://doi.org/10.1016/j.bspc.2016.11.004
11. Malik, K.R., Sam, Y., Hussain, M., Abuarqoub, A.: A methodology for real-time data
sustainability in smart city: towards inferencing and analytics for big-data. Sustain. Cities
Soc. 39(April), 548–556 (2018). https://doi.org/10.1016/j.scs.2017.11.031
12. Ye, Y., Zhao, Y., Shang, J., Zhang, L.: A hybrid IT framework for identifying high-quality
physicians using big data analytics. Int. J. Inf. Manag. 47(January), 65–75 (2019). https://
doi.org/10.1016/j.ijinfomgt.2019.01.005
13. Allam, Z., Dhunny, Z.A.: On big data, artificial intelligence and smart cities. Cities 89
(January), 80–91 (2019). https://doi.org/10.1016/j.cities.2019.01.032
14. Zhang, L., Wen, J.: A systematic feature selection procedure for short-term data-driven
building energy forecasting model development. Energy Build. 183, 428–442 (2019). https://
doi.org/10.1016/j.enbuild.2018.11.010
15. Wang, F., Liang, J.: An efficient feature selection algorithm for hybrid data. Neurocomputing
193, 33–41 (2016). https://doi.org/10.1016/j.neucom.2016.01.056
16. Lin, Y., Wang, H., Zhang, S., Li, J., Gao, H.: Efficient quality-driven source selection from
massive data sources. J. Syst. Softw. 118, 221–233 (2016). https://doi.org/10.1016/j.jss.
2016.05.026
17. Manbari, Z., AkhlaghianTab, F., Salavati, C.: Hybrid fast unsupervised feature selection for
high-dimensional data. Expert Syst. Appl. 124, 97–118 (2019). https://doi.org/10.1016/j.
eswa.2019.01.016
18. Mahdiyah, U., Irawan, M.I., Imah, E.M.: Integrating data selection and extreme learning
machine for imbalanced data. Proc. Comput. Sci. 59(ICCSCI), 221–229 (2015). https://doi.
org/10.1016/j.procs.2015.07.561
19. Harinarayan, V., Rajaraman, A., Ullman, J.D.: Implementing data cubes efficiently.
SIGMOD (1996). https://doi.org/10.1145/235968.233333
20. Peña-Ayala, A.: Educational data mining: a survey and a data mining-based analysis of
recent works. Expert Syst. Appl. 41(4 PART 1), 1432–1462 (2014). https://doi.org/10.1016/
j.eswa.2013.08.042
21. Romero, C., Ventura, S.: Educational data mining: a survey from 1995 to 2005. Expert Syst.
Appl. 33(1), 135–146 (2007). https://doi.org/10.1016/j.eswa.2006.04.005
22. Slater, S., Joksimovic, S., Kovanovic, V., Baker, R., Gasevic, D.: Tools for educational data
mining: a review. J. Educ. Behav. Stat. 42(1), 85–106 (2017). https://doi.org/10.3102/
1076998616666808
23. Han, J., Pei, J., Dong, G., Wang, K.: Efficient computation of Iceberg cubes with complex
measures, In: Proceedings of the ACM SIGMOD International Conference on Management
of Data, vol. 30, no. 2, pp. 1–12 (2005). http://www.scopus.com/inward/citedby.url?scp=
0034825777&partnerID=8YFLogxK
Intelligent Learning Ecosystem in M-Learning
Systems
Diana Carolina Burbano Gonzalez1,

Clara Lucia Burbano Gonzalez2(&), and Cristian Barria Huidobro3
1
Universidad San Buenaventura, Cali-Valle, Colombia
caritoburg@yahoo.com
2
Corporación Universitaria Unicomfacauca, Popayán-Cauca, Colombia
cburbano@unicomfacauca.edu.co
3
Universidad Mayor, Santiago, Chile
cristian.barrio@mayor.cl
Abstract. This article refers to the background of Education and educational

technology; To understand, they are theoretical-explanatory links in the
teaching-learning process in today’s networked society. The configuration of the
learning environment and the development of experiences that generate learning
by integrating the monitoring of real-time experiences into ecosystems that
guide the learning outcomes common to all and desirable. The epistemological
approach is an “alternative” proposed by [1], it facilitates the construction of
new knowledge and the method is a research action of analysis the exchange of
mixed studies mediated by mHealth in the thematic of the pathological oral
anatomy. Data were obtained through focus groups to students; The data
analysis was performed in MaxQd v 2018 from the analysis categories. The
results are organized according to the analysis categories for students, a saber:
dialogue; interaction with teachers, interaction with peers, perceived learning;
Mediation with applications.
Keywords: Digital learning ecosystem Intelligent learning Learning

interaction and m-learning
1 Introduction
The current information society poses an environment in which ICTs play a strong role
in all areas, including education [2]. This demands the establishment of pedagogical
models aimed at both the design of educational materials in digital format and their
mode of application [3] In this sense, higher education institutions have been forced to
adopt strategies to respond to new trends in information and communication by com-
peting with a greater advantage among their peers; by offering alternatives to access
knowledge different from traditional forms, which require the convergence of teacher
and student time-space. The use of information and communication technologies
(ICT) offers a wide variety of tools in the educational process, consequently, by
themselves, they do not guarantee appropriate learning, which is why the importance of
pedagogical strategies that support the learning of learning is evident. Collaborative way

https://doi.org/10.1007/978-3-030-33229-7_19
214 D. C. Burbano Gonzalez et al.
Identifying student learning at any time is important, on the one hand, because of prior
learning conditions the construction of new learning and, on the other hand, because of
monitoring, both by the same apprentice and by whoever oriental is a fundamental
condition for its control and direction. This dimension of research requires the appli-
cation of measurement methods and statistical procedures [4].
The use of intelligent learning systems with analytical methods in online education
focuses on the collection, analysis, and reporting of data on student online actions [5].
The learning analysis has aroused the interest of academic institutions and teachers by
displaying information that was not available before, allowing them to better under-
stand how students learn and, therefore, to take actions with information to support the
process [6]. Advances in technology in recent years have changed student learning
behaviors and teaching methods with new environments and ubiquitous resources
methodologies. Consequently, our goal is to provide a theoretical and analytical
understanding of the pedagogical model and its implications in teaching and learning.
The analysis of the pedagogical model and didactic sequence (thematic of basic sci-
ences in oral pathology anatomy) is implemented by generating interactions with
students in the new m-learning environments, which integrate students into learning
outcomes common to all and desirable. The m-learning education allows both teachers
and students to connect through their mobile devices generating communicative
interaction.
This article reveals how technological mediation is generated in learning, from a
trans and interdisciplinary perspective, by establishing relationships between knowl-
edge. This is possible because the authors root their discourse in an epistemic frame-
work and an objective that makes all knowledge common, through reflexive and
dialogical analysis.
2 State of the Matter
The technology has offered a line of advancement to the attention in intelligent learning
and the creation and efficient use of intelligent learning environments mediated with
ubiquitous technology.
2.1 Ubiquitous Learning and Intelligent Learning

The rapid advance of the networks of communication wifi, 4 g, 5G .., and the use of the
Smartphone have allowed the students to access digital resources and interact with
computer systems without being limited by location and time. This has been investi-
gated in this perspective by [7], and with criteria based on learning by [8]; The authors
call this learning approach that uses mobile and wireless communication technologies,
“mobile learning” [9] or “ubiquitous learning” [8]. The other pillar is the detection
technologies: GPS (global positioning systems), RFID (radio frequency identification),
QR (rapid response); they have allowed the learning systems to detect the real-world
locations and contexts of the students [10] and the recommendation systems [11, 12].
Various studies have evidenced these approaches and have emphasized the rele-
vance that students learn from the real world with access to digital resources they use in
Intelligent Learning Ecosystem in M-Learning Systems 215
their activities and are assisted by their teachers in technology-mediated learning and
web-based learning, towards mobile learning and especially “ubiquitous context-aware
learning” [13].
2.2 Reference Framework of the McLuhan Tetrad

McLuhan Media Laws are based on the idea that all artifacts have an effect on people
and the society that adopts them. From this perspective, each new tool that is intro-
duced becomes an extension. The word “extension” refers to the idea that, by building
new things, humans increase their bodies and these changes are, in the long term,
transforming the social and physical environment. We need an answer to a fact, the
transit of the intelligent university (intelligent learning systems); through analysis based
on experiences and interaction in learning ecosystems (m-Learning).
The data sources used in the learning analysis tools can come from a variety of
academic systems, such as student information, library services, learning management
systems, student admissions and grades [5]. Among those sources, research on learning
analysis has tended to focus on the possibilities of using data in LMS learning man-
agement systems. According to Vernet, the potential uses of learning analysis are
related to the following areas: (1) prediction of student performance and student
modeling, (2) suggestion of relevant learning resources, (3) increased reflection and
awareness, (4) improvement of social learning environments, (5) detection of unde-
sirable learning behaviors and (6) identification of student emotions.
The research field of learning analytics is interdisciplinary, consequently, it com-
bines aspects of educational data mining, social network analysis, artificial intelligence,
psychology - theory and educational practice. In educational research, it is important to
define two related areas: learning practice and organizational development. Both use
educational data, although with different interests. While, in the practice of learning, the
analyzes focus on improving student success, in organizational development the
pressure has been on productivity and business solutions. In the domain of the orga-
nization, the analysis combines student information with institutional data to improve
administrative efficiency [14].
The literature on learning analysis tools also suggests that it can be used to address
data from a variety of perspectives. Some of the most prominent are social networks,
discourse analysis [15, 16], content analysis [17, 18], disposition analysis [19] and
student-centered analysis [20], among others. In all of them, learning analysis tools are
expected to improve teaching and support student success.
2.3 Cognitive Sciences

From the cognitive sciences, including neurosciences and their relationship with psy-
chology, anthropology and information sciences, innumerable contributions are made to
the understanding of m-learning teaching and learning. How people learn using digital
media, how information processing is carried out, how perceptual, associative and inte-
grative processes carried out in the brain are involved, what are the best ways to teach from
the functioning of the nervous system, or As meanings are built and skills are acquired,
they are some of the questions to which cognitive sciences can contribute to answer.
2.4 Learning Analysis Tools

There are multiple definitions of the concept of “learning,” and these depend on the
epidemiological position on which it is based. For this article, the conceptualized one of
[21] has been selected, who sees learning as A persistent change in human performance
or its potential for performance, which must be the result of the subject’s experience
and interaction with the world
Based on the McLuhan axiom, all media are extensions of people, we analyze how
the tools of learning analysis broaden our senses as human beings. Learning analysis
tools reveal “hidden” information, it connects with the idea of a sixth sense, in this case
to perceive learning behaviors that are not visible. By displaying this data, the learning
analysis allows another view of what happens when students participate in online
learning. The expectations of the learning analysis go beyond having a different view
on teaching and learning. In this sense, education professionals and academics have
expressed their hopes in the learning analysis by predicting learning performance and
identifying learning models. Customize learning, control the activity of teachers, as
well as the performance of the institution, understand social interaction and partici-
pation and engage students in their learning processes.
The prediction of student success or failure in learning, particularly in e-learning,
has received considerable attention. Research in this area has led to the definition of
profiles to model different types of learners, as well as the identification of different
learning styles. Personalization is the core of many approaches to learning analysis;
The data collected on students’ online behaviors inform decisions about what type of
learning resources or activities are most significant, given the student’s current skills
and knowledge on a given topic. Taking into account individual aspects, the tools of
learning analysis improve a broader vision of learning that recognizes the importance
of building prior knowledge and skills of students.
Learning analysis tools seek personalization, are based on the idea of how students
learn, therefore, to ensure that students acquire the desired skills, teaching practice must
adapt to the diversity of needs and challenges faced by students.
In many learning analysis tools, in addition to individual performance, data on
group activity is also available. This feature intensifies the comparison between the
individual and the group and indirectly pushes students to work harder when their
activity falls below the group average. Educational institutions have used learning
analysis to recruit students [22] and sometimes speculate on the possibility of human
resources in the analysis of future learning. This scenario forces people to compete to
guarantee access to the university or the labor market.
Nowadays societies need collaboration and cooperation instead of competition, the
idea that educational institutions need to prepare students to work in a competitive
society has been described as myth [23]. One of the main effects of Competition is
homogenization: people need to share the same objectives and rules to compete [23]. In
learning, standardization implies that everyone must learn the same in the same period.
Continuous monitoring and pressure to meet academic expectations can create anxiety
and distrust among those interested in education. [24] has studied these phenomena in
learning in the workplace, and has concluded; in the monitoring of the online learning
activity negatively affects the collaboration of workers, communication and knowledge
sharing. Although formal education differs from learning in the workplace, the stress on
students caused by fulfilling their curriculum on time while staying at the same level as
the group does not support creativity and innovation [24]. Therefore, to the extent that
the tools of learning analysis do not recognize the value of experimentation and risk-
taking, they intensify a vision of learning based on the efficiency in which failure is
penalized.
Digital data has become a key element in management techniques that are
“evidence-based.” Educational institutions are subject to a logic similar to that of
contemporary organizations, which are based on the use of data and information.
A good example of this trend can be found in the university and the school, in which
the emphasis on the indicators has been questioned, therefore hiding good practices in
teaching and learning. From this perspective, the most critical voices They affirm “the
analysis of learning, especially the academic analysis, intensifies the management
culture in education” [25], some sectors of the academic community affirm that the
analyzes can improve the understanding and performance of the students [26] For
teachers, the possibility of accessing the data generated by the students allows them to
reflect on the ins design and management of the courses they teach. In this case, the
learning analysis is presented as a tool that promotes the knowledge and reflection of
educators about some aspects of their professional practice.
Learning analysis is part of a trend based on decision-making informed by data.
From this perspective, automatic collection and analysis of student behavior data are
assumed to be relevant and reliable, or at least more reliable than subjective percep-
tions. The trust dedicated to computer algorithms is not exclusive to learning analysis,
and similar attitudes towards data in business, medical care, social services, sports, etc.
can be found. Although stakeholders in education recognize that learning analysis
enriches teaching and learning, we could question the extent to which learning analysis
is affecting the credibility we give to personal impressions. In a data-driven society, can
we rely on subjective and qualitative data collected through individual experiences?
The learning analysis modifies certain aspects of the role of teachers, especially in
online education. Here, we could say that the confidence in the data of the learning
analysis is closely related to the emergence of educational programs online. As Mazza
and [27] point out, in e-learning courses, students face related challenges such as
loneliness, technical problems or loss of motivation [28]. In these cases, the lack of
visual guidelines from teachers that help them recognize when students are poorly
motivated, anxious or overwhelmed is compensated through learning analysis. In
mixed learning scenarios, the learning analysis affects the teacher’s ability to perceive
group feedback, as there is a growing tendency to rely on the information collected
through the back channels during large conferences. The ultimate goal of these efforts
is to improve adaptation and improve teaching. But as McLuhan and McLuhan [29]
pointed out, the simultaneous effect is the disappearance of certain practices. In this
case, the praxis that is being relegated is the ability of certain teachers to detect
individual and group behaviors.
2.5 Interactions
The pedagogical design currently proposed by the traditional education system requires
the adequacy of the design of competent and individualistic producing minds that
accommodate an industrialized social context. This learning structure based on the
mass production paradigm seeks to retake the “feed-feed” model within the educational
processes; a model that uses different pedagogical practices of collaborative interaction
that involve technological and virtual communication scenarios [30].
This type of learning scenarios generally considered communication devices are
centralized in the development of participatory cognitive structures within the social
and educational environment, with the support of convergent interactive technologies
to articulated languages for different purposes [31]. This technological accompaniment
from the constant interactivity of individuals within a currently techno-mediated
environment [32], arises amid the need to create new collaborative relationships of
dissimilar transmissions.
The manifestation of these transmissions based on interaction relationships forms a
connective knowledge based on autonomy, diversity, openness and
connectivity/interactivity [31]. For Badillo, these interactions arise from the dialogue
and argumentation of individuals, which allows the reciprocal exchange of different
opinions, perspectives, and reconstruction of meanings, where new technologies are
fundamental in collective interaction, differentiators in the reorganization of mental
functions; they assign the meaningful concept of learning and enable the basic principle
of knowledge transfer to real situations [31].
2.6 Mobile Learning

The new century is going through an era where information and knowledge are
accessible from anywhere and at any time. It can be said that this is a society saturated
with information influenced by technology and science available to the world. A so-
ciety that is characterized by the variety of autonomous contexts differentiated by
resources, willing to take on technological innovation challenges to improve their
development, it can be said that this society is going through the phenomenon of
ubiquitous training [33]. The current society adapts to the new mediated techno
ecosystem [32], must emerge from its traditional roots to establish an educational
system that integrates new models of quality, knowledge, and resources [34].
This is a challenge that the current educational model must take with care and
responsibility, which goes through an outdated stage that refers to the beginning of its
development before the emergence of information and communication technologies
[34]. The current model, based on the face-to-face design of traditionalist practices
[31], is aimed at educating a certain segment of the population. Therefore, it is
important that training begins to be restored from the educational model of teachers;
adapt them to the modern education system that includes technology among its
practices.
We must not forget the work that these devices play in formal education since they
are considered excellent administrators of academic material. They provide valuable
and interactive multimedia learning content for educational purposes [35]. Also,
learning strategies appropriate to this context can help educators facilitate the mobile
learning process and reach their educational goals. Likewise, when these materials are
properly adapted to their educational formats, the student can take full advantage of
spaces that until now were not part of the learning time [36].
It is necessary to analyze the options for adapting training content (which are
generally considered robust) to mobility scenarios, that is, transferring learning
methods to a mobile context-based on flexible and interactive styles that are not dif-
ficult to grasp [37]. The use of technology alone does not directly include improve-
ments in training. Cañizares points out through the author [38], artificial intelligence
theorist and learning expert, the skepticism about a large part of the courses offered
online; It considers that there is no sample of changes in the educational models and
they are conditioned to offer the same materials as always. Schank is committed to
learning based on experience. As a student of the human mind, he states that mental
processes evolve greatly from experience, given that he faces real situations. Contrary
to what happens with conventional learning, based on questions and answers outside
the real context and far from the individual’s praxis [36].
Education should examine how educational resources are designed and delivered
and take into account the needs and characteristics of the new student generation. For
example, in the delivery of improved technology to the student, what is the purpose of
the tool at the educational level? Does it fit the current training needs? The current
generation of students uses this technology in continuous transformation, such as
mobile devices, which require the support of teachers as well as information and timely
feedback. It must be taken into account that these changes occur according to the
training needs supported by current technologies that arise in a society with high
information demand [39]. The educational system must rescue informal learning
practices adapted to the new connective and flexible world in the use of mobile
technology, which is aimed at the autonomous and collaborative learning of students.
Without a doubt, this is a knowledge society that must be willing to take on the
challenges associated with communicative and informational technologies [40].
3 Research Design
Information and communication technologies (ICT) represent each, new challenges and
possibilities in the educational field. The term educational technology acquires a new
meaning that refers to the incorporation of ICT in teaching and learning processes. This
emerging field called technological mediations in education supposes its epistemo-
logical status, supported by positioned epistemic tendencies, which reveal cognitive
configurations that focus its interest in categories such as learning and collaboration,
privileging interactivity, under the assumption that it gives meaning to experience and
applicability of the contents learned to real-life situations.
Following the reflections of [41], it is possible to overcome the false dichotomy
between quantitative research and qualitative research in the social and human sci-
ences, for which he postulates an integrative vision called “alternative research”. “The
two types of techniques need each other most of the time … every existence or
phenomenon has quantitative and qualitative attributes … Consequently, some authors

propose overcoming this false dichotomy.”
4 Research Method
The Research-Action (I-A) Lewin’s work in the period immediately after World War II.
Lewin identified four phases in I - A (planning, acting, observing and reflecting) and
imagined it based on the principles that could lead “gradually towards independence,
equality, and cooperation” [42]. Throughout these years, the I-A method has been
configured based on numerous contributions from different geographical and ideo-
logical contexts. The great diversity of conceptions that currently exist around I-A, both
from a theoretical and experiential perspective, makes it less than impossible to arrive
at a unique conceptualization. However, there are many common features in which
most authors are coincidental. In the first place, it is worth highlighting the prepon-
derant nature of the action, as defining this research method. This dimension is spec-
ified in the active role assumed by the subjects who participate in the research, which
takes as a start the problems arising from educational practice, reflecting on them,
breaking in this way with the separatist dichotomy theory/practice. “Action research is
a form of research carried out by practitioners on their practices” [43].
5 Hypothesis
There is a significant transformation of the preconceptions that, on oral pathology,

periodontics students bring, under the use of the mHealth application.
6 Techniques and Instruments for Information Gathering
In the present investigation, the following instruments will be used for the collection of
the information: script of questions for focus group (qualitative analysis) and the
questionnaire (pre-test) in the mobile application (quantitative analysis).
7 Population and Sample
The target population is heterogeneous and is made up of the academic program of

Specialization in Periodontics and Osseointegration existing in the Cooperative
University of Colombia, Bogotá headquarters; attached to the Faculty of Dentistry,
which has informed consent or voluntary acceptance of participation in the project. The
number of residents 15, enrolled according to Helmsman in the second and fourth
semester in the second half of 2018.
8 Consent to Participate in the Study
In line with internationally accepted standards to ensure the ethical treatment of people
involved in scientific research, an informed consent form was developed. The document
will explain to the research participants in the focus groups the nature of the study, its
objectives, as well as the right not to participate or to suspend their participation at any
time and the confidentiality of the information provided to the participants.
9 Data Analysis Plan
The proposed methodology is complementary (quantitative and qualitative), the anal-

ysis of the data obtained according to their nature will be performed. For the processing
and analysis of quantitative data, the statistical software SPSS (statistical package for
the social science), version 20 for Windows, was used. As recommended by various
authors such as [44], first the descriptive statistics will be kept, including frequencies,
calculation of measures of central tendency (mean, median and mode) and dissension
(variance and standard deviation). The analysis is performed between differences of
groups (T students test for independent samples and analysis of variance of a factor)
and the degree of association or relationship between the variables (Pearson’s corre-
lation coefficient and chi-square tests).
The qualitative information collected will be analyzed with the MaxQda software,
version 2018. The interviews are made recorded in audio format and subsequently
transcribed and incorporated into the software for the organization and classification of
data, summarize and tabulate them (data reduction). It will be done following the
inductive method, based on the information obtained, the analysis categories will arise.
Therefore, following the suggested methodology, qualitative data will pass to an
analytical description to extract relevant information, from coding to the interpretation
of its meaning and importance [45].
The preparation of qualitative data includes: The transcription of interview
recordings, observation and field notes (computer programs for optical recognition of
OCR characters and voice recognition).
Thematic coding and categorization: coding is a way of indexing or categorizing
the text to establish a framework of thematic ideas.
Narratives: storytelling or storytelling is one of the fundamental ways people
organize the understanding of the world. In the stories, people make sense of their
experience and share their experience with others. Therefore, careful analysis of the
themes, content, style, context, and narrative reveals people’s understanding of the
meaning of events in cultural contexts.
Comparative analysis: the coding provides the shorthand synthesis between dif-
ferent people, objects, scenes or events (situations, actions, stories or experiences of the
members), data of the same people, scenes, objects or types of events, some matches
with others.
Typology: a typology is a way of classifying things that can be multidimensional or
multifactorial, can be based on two (or more) different categories of things.
Quality of analysis and ethics
10 Data Analysis of Focus Groups with Qualitative Analysis

Techniques
In the present study the classic content analysis will be used, it implies the creation of
small pieces of data (Chunks) and the subsequent assignment of a code to each one of
them. However, instead of creating a theme from the codes, these are placed in similar
groups and then accounted for. Using the three-element coding framework there are
only three ways to use classical content analysis with focus group data: a. the analysis
can establish if each participant uses a given code; b. the analyst can determine if each
group used a given code; c. the analyst can identify all instances in which a certain code
appears [46]. Researchers should not only provide information on the frequency of
each code (that is, quantitative information) but also complement it with a description
of each code (qualitative information); which will create a mixed modality of content
analysis [47].
11 Analysis of Quantitative Data
11.1 Interpretation of Pre Test Results

Regarding the content of m-learning educational applications by students, the results
show a 30% difference between women and men. 40% of women contain educational
applications on their mobile devices, unlike 70% of men. However, the data show the
lack of good use of the mobile device in educational contexts by men, because women
predominate communicative and interactive participation in different learning
environments.
Although the methodology used by the teacher has shown good results, it is
important to clarify that from the technological - pedagogical perspective, there is a
need to include information and communication technologies (ICT) in their training
processes. The study shows low rates about the use of the mobile device in the edu-
cational field, as well as the management of mobile educational applications. Therefore,
the support of ICT, specifically mobile devices (m-learning) for the teaching-learning
process, is undoubtedly the opportunity to transform formal education into a mobile
and ubiquitous context applied to the reinforcement of knowledge in the subject of
pathological oral anatomy.
11.2 Post-test Analysis

Once the pedagogical - technological and communication model has been imple-
mented, the interaction - gender variable continues to stand out the intervention of
women in terms of conditions of a teacher-student relationship, synchronous and
asynchronous technological interaction), active participation in the subject, and the
Access to resources on the platform.
The analysis of the comparative statistical results of the pre-test and post-test shows
in the study of interactions in collaborative learning environments, mediated by m-
learning, a significant difference is reported in the present comparative study. Figure 1

shows a before and after the m-learning test between the two samples.
Fig. 1. Pre test and post test behavior. Source own elaboration
The behavior of the graph demonstrates the changes obtained once technological
mediation has been applied in periodontics residents, due to the fact that a high fre-
quency of technological and social interaction is evident. The average score that
marked the difference between pre and post was 7.86.
Technique: T-Student Test for Related Simples
The t-student test was applied, to calculate and know the values of the mean, standard
deviation and the standard mean error of the sample, as shown in Table 1.
Table 1 Paired samples 1. Source own elaboration

Paired sample statistics
Half N Standard deviation Mean standard error
Pair 1 INTERACTION (PRE) 27,40 15 4,837 1,249
INTERACTION (POS) 35,27 15 5,244 1,354
The results show a difference of the average between the pre and post-test at a 7.867
higher in the result of the evaluation of the interactions that the post-test throws. For the
standard deviation, the dispersion of the data is more significant in the post-test with a
score of 5,244 compared to 4,837 that produced the pre-test; This is due to atypical data
presented within the post-test results.
Regarding Pearson’s linear correlation (Table 1), in this case, it is 0.100. As this
value of significance is greater than the value of alpha (0.05), it can be statistically
deduced that the scores in both registration conditions do not correlate, that is, there are
differences in their results.

Matched sample correlations
N Correlation Sig.
Pair 1 INTERACTION (PRE) & INTERACTION (POS) 15 ,440 ,100
To know if there is a significant statistical difference in the results obtained, the

difference between the P-value (0.000) and the value of alpha (0.05) was calculated. In
this case, since P-value is less than alpha, it can be deduced that the statistical dif-
ference between the tests is significant (Table 2). Another condition for complying with
this theory is shown in the relationship between the lower limit and the upper limit of
the confidence interval is not 0 (Table 3).

Test paired samples
Half Standard Mean error 95% difference
deviation are-given confidence interval
Pair 1 INTERACTION (PRE) - −7,867 5,343 1,380 −10,826
INTERACTION (POS)
95% difference confidence interval
Higher t gl Sig(bilateral)
Pair 1 INTERACTION (PRE) - −4,908 −5,702 14 ,000
INTERACTION (POS)
It can be concluded that there is a significant difference in the means of the

interaction variable before and after the technological mediation because P-value (level
of significance) is less than 0.05 (alpha), so the test m -learning has significant effects
on student interaction.
11.3 Qualitative Analysis

In qualitative research, the categories of analysis are dynamic and are built throughout
the process, therefore general categories are established, derived from the research
question. The initial proposal of analysis categories was based on theoretical revision
and conceptual construction for the study.
11.4 Data Analysis

Concerning the interpretation and discussion of the data, it was done by contrasting the
experiences of students and teachers with m-learning, dialogue, interaction, and
learning. The information collected was analyzed based on categories with the help of
the MaxQA 2018 software program. The method includes data reduction, coding,
theoretical sampling. The principle of qualitative information analysis was that of
“saturation”, thus the cyclical analysis of the information comparing data to form
categories, relating them to each other and resorting to new data and informants to
confirm, modify or reject them, special attention was given to even a saturation point,
that is, the new cases will not provide new information, which indicated that at that
time the analytical process should conclude.
11.5 Interpretation of Qualitative Results

Analysis Category: The function of the dialogue with classmates and teachers
The dialogue functions are congruent with the predominant use of discussion
forums. in m-learning, when conducting debates, exchanging perspectives among
students and raising doubts with teachers. The possibility of contrasting opinions and
interacting with other people with different experiences is positively valued by stu-
dents, as it contributes to the development of professional skills.
The characteristics of the dialogue with codes were: communication is very
important, respect and trust in the interaction, dialogue with peers are enriching, large
and heterogeneous groups generate communication, communication with peers is fluid,
communication with the teacher is fluid and generates feedback.
In general, the dialogue is perceived as relevant, respectful and enriching; High-
lights the recognition by students of the importance of communication, as well as the
elements of respect and trust, which are part of the very concept of dialogue. On these
aspects, in a dialogical relationship, the student affirmed:
“Dialogue in apss-mediated learning, with respect and confidence in the fulfillment
of acquired commitments.”
Interaction with Teachers: What is the degree of interactivity between students and
teachers? What is the quality of the interaction? And How does it take place?
The analysis of this category, derived from the focus groups, was broken down into
four sets of codes: tools used to interact with teachers; functions of interaction with
teachers, frequency of interaction with teachers, quality of interaction with teachers.
About how interaction with teachers takes place, the most frequently used media were
the asynchronous discussion forums, as well as email.
Interaction with Students: How is dialogue between students encouraged? and what
role does it play in the student’s learning process?
The analysis of the information provided by students in the focus groups resulted in
code groups: tools used when interacting with peers, frequency of interaction, functions
of interaction with peers, quality of interaction with peers. Then a representative
appointment of the forms of communication used: “We communicate through tele-
phone messages and emails. Communication through electronic messages and emails is
given to share opinions about the course and organize better group work, interact
constantly, even more than in the classroom.
12 Discussion of Results from the Conceptual Framework
Dialogue. In terms of dialogue, the virtual courses analyzed, from the perspective of
the students, have an appropriate level of dialogue, understood as a bidirectional,
horizontal, respectful and learning-centered relationship, between students and teachers
and between the students themselves. The intensity of the dialogue is variable, although
none of the courses presented a major difficulty that prevented the learning process.
The availability of technological tools allows an acceptable level of dialogue.
Structure. Students perceive that the courses have an appropriate level of structure:
course program, reading material, support material, links and activities scheduled
according to a schedule. However, sometimes the lack of structure is a factor that
negatively affects students’ perception of teacher performance and the degree of per-
ceived learning.
Direct instruction is exercised by teachers asynchronously, through a “mediated
dialogue”, as previously seen; They use written materials available on the platform, as
well as announcements and messages to intervene in the presentation of content. How-
ever, students must assume their understanding of the material by reading and evacuating
doubts at the time required, mainly with the use of the question forums and email.
13 Conclusions
A current trend in the “teaching-learning system”, as opposed to the pedagogical

practice of the teacher; Currently, teaching is less interesting than learning, and less
interesting what the teacher presents and more what the student learns (training focused
on learning).
It was found that students generally demonstrated an adequate level of satisfaction
in m-learning, by separating the analysis between the best-rated courses was the dis-
cussion forum, the exchange of opinions and experiences with classmates were
assessed as teaching-learning strategies.
Learning in collaborative or cooperative environments allows greater reception in
educational settings. The student these learning options is in the ability to make
decisions about their process and identify their own needs, specifically in the subject of
networks which presents a high degree of student dropout.
Collaborative learning allows each student to develop their creative thinking, self-
learning, commitment and responsibility of their peers, and their participatory spirit,
achieving group growth.
The present investigation demonstrated the effectiveness of peer interaction gen-
erates skills in the subject of networks and stimulates cognitive improvement; At the
time of exchange, we establish a relationship between the interacting subjects, moti-
vations, and interests.
Interaction with peers has the function of coordinating tasks in group work or
projects, as well as exchanging opinions and learning from the experiences of others.
The degree of interaction is adequate but depends largely on the design of the activities.
The teacher’s role remains important as a promoter of the interaction in the group.
The main contribution of this research is the approach of three elements directly
related to the positive perception of pedagogical presence by students, namely: feed-
back, academy, and socialization. In general, both students and teachers consider the
contribution that peer interaction makes learning as important.
References
1. Paramo, P.: La investigacion en ciencias sociales: estrategias de investigacion. Universidad
piloto de Colombia, Bogota, p. 332 (2003)
2. Galvis, E., Sánchez, J.: A critical review of knowledge management in software process
reference models. J. Inf. Syst. Technol. Manag. 10 (2013). https://doi.org/10.4301/S1807-
17752013000200008
3. Rosanigo, Z., Bramati, P.: Objetos de aprendizaje. XIII Workshop de Investigadores en
Ciencias de la Computación, pp. 574–869 (2011). http://sedici.unlp.edu.ar/bitstream/handle/
10915/19934/Documento_completo.pdf?sequence=1. Accessed 7 Sept 2013
4. Maldonado, L.: Virtualidad y autonomía pedagogía para la equidad. ICONK Editorial,
Colombia (2012)
5. Lockyer, L., Heathcote, E., Dawson, S.: Informing pedagogical action: aligning learning
analytics with learning design. Am. Behav. Sci. 57(10), 1439–1459 (2013)
6. Dawson, S., Gašević, D., Siemens, G., Joksimovic, S.: Current state and future trends: a
citation network analysis of the learning analytics field. In: Proceedings of the Fourth
International Conference on Learning Analytics and Knowledge, March 2014, pp. 231–240.
ACM (2014)
7. Hwang, G., Chang, H.: A formative assessment-based mobile learning approach to
improving the learning attitudes and achievements of students. Comput. Edu. 56, 1023–1031
(2011)
8. Zapata-Ros, M.: Teorías y modelos sobre el aprendizaje en entornos conectados y ubicuos.
Bases para un nuevo modelo teórico a partir de una visión crítica del “conectivismo”. Teoría
de la Educación. Educación y Cultura en la Sociedad de la Información, vol. 16 (2015)
9. Sharples, M., Arnedillo-Sánchez, I., Milrad, M., Vavoula, G.: Mobile learning. In: Balacheff,
N., Ludvigsen, S., de Jong, T., Lazonder, A., Barnes, S. (eds.) Technology-Enhanced
Learning, pp. 233–249. Springer, Dordrecht (2009). https://doi.org/10.1007/978-1-4020-
9827-7_14
10. Hwang, G., Furuchi, T., Naganuma, A.: The ubiquitin-conjugating enzymes, Ubc4 and
Cdc34, mediate cadmium resistance in budding yeast through different mechanisms. Life
Sci. 82(1182–5), 23–24 (2008)
11. Pazzani, M., Billsus, D.: Content-based recommendation systems. In: Brusilovsky, P.,
Kobsa, A., Nejdl, W. (eds.) The Adaptive Web. LNCS, vol. 4321, pp. 325–341. Springer,
Berlin (2007). https://doi.org/10.1007/978-3-540-72079-9_10
12. Chen, Y., Cheng, L., Chuang, C.: A group recommendation system with consideration of
interactions among group members. Expert Syst. Appl. 34(3), 2082–2090 (2008)
13. Liu, G., Hwang, G.: A key step to understanding paradigm shifts in e-learning: towards
context-aware ubiquitous learning. Br. J. Edu. Technol. 41, E1–E9 (2010)
14. Siemens, G., Long, P.: Penetrating the fog: analytics in learning and education. Educause
Rev. 46(5), 30 (2011)
15. De Liddo, A., Shum, S., Quinto, I., Bachler, M., Cannavacciuolo, L.: Discourse-centric
learning analytics. In: Proceedings of the 1st International Conference on Learning Analytics
and Knowledge, pp. 23–33 (2011). ACM
16. Ferguson, R., Shum, S.: Learning analytics to identify exploratory dialogue within
synchronous text chat. In: Proceedings of the 1st International Conference on Learning
Analytics and Knowledge, pp. 99–103. ACM (2011)
17. Drachsler, H., Bogers, T., Vuorikari, R., Verbert, K., Duval, E., Manouselis, N.: Issues and
considerations regarding sharable data sets for recommender systems in technology
enhanced learning. Procedia Comput. Sci. 1(2), 2849–2858 (2010)
18. Verbert, K., Drachsler, H., Manouselis, N., Wolpers, M., Vuorikari, R., Duval, E.: Dataset-
driven research for improving recommender systems for learning. In: Proceedings of the 1st
International Conference on Learning Analytics and Knowledge, pp. 44–53. ACM (2014)
19. Crick, R., Broadfoot, P., Claxton, G.: Developing an effective lifelong learning inventory:
the ELLI project. Assess. Edu.: Principles Policy Pract. 11(3), 247–272 (2004)
20. Kruse, A., Pongsajapan, R.: Student-centered learning analytics. CNDLS Thought Papers,
1–9 (2012)
21. Driscoll, M.: Psychology of Learning for Instruction. Allyn & Bacon, Needham Heights
(2000)
22. Van Harmelen, M., Workman, D.: Analytics for learning and teaching. CETIS Analytics Ser.
1(3), 1–40 (2012)
23. Combs, A.: Myths in Education: Beliefs that Hinder Progress and their Alternatives. Allyn
and Bacon, Boston (1979)
24. Wesley, D.: A critical analysis on the evolution of e-learning. Int. J. E-learn. 1(4), 41–48
(2002)
25. Selwyn, N.: Data entry: Towards the critical study of digital data and education. Learn.
Media Technol. 40, 64–82 (2014)
26. Graf, S., Ives, C., Rahman, N., Ferri, A.: A tool for accessing and analysing students’
behaviour data in learning systems. In: Proceedings of the 1st International Conference on
Learning Analytics and Knowledge, pp. 174–179. ACM (2011)
27. Mazza, R., Dimitrova, V.: Visualising student tracking data to support instructors in web-
based distance education. In: Proceedings of the 13th International World Wide Web
conference on Alternate track papers & posters, pp. 154–161. ACM (2004)
28. McLuhan, M.: Understanding Media. The Extensions of $man. Sphere Books, London
(1964)
29. McLuhan, M., McLuhan, E.: Laws of Media: The New Science, vol. 1. University of
Toronto Press, Toronto (1988)
30. Aparici, R., Silva, M.: Pedagogía de la interactividad. Comunicar: Revista Científica
Iberoamericana 19(38), 51–58 (2012). https://doi.org/10.3916/C38-2011-02-05
31. Badillo, E.: Tensiones Comunicativas Emergentes En Estrategias Colaborativo. Red de
Revistas Científicas de América Latina, El Caribe, España Y Portugal, vol. 9, pp. 188–201.
http://www.redalyc.org/articulo.oa?id=265428385012
32. Ortiz, R., Ramirez, A.: Ellos vienen con el chip incorporado. Editorial, Bogotá
33. Gros, B., Maina, M.: The Future of Ubiquitous Learning. https://doi.org/10.1007/978-3-662-
47724-3. Barcelona
34. Ramírez, M.: Modelos y estrategias de enseñanza para ambientes innovadores. In: Modelos
y Estrategias de Enseñanza para ambientes innovadores, pp. 1–55 (2013). https://www.
editorialdigitaltec.com/materialadicional/ID254_RamirezMontoya_
Modelosyestrategiasdeensenanza.cap1.pdf
35. Jeng, Y., Wu, T., Huang, Y., Tan, Q., Yang, S.: The add-on impact of mobile applications in
learning strategies: a review study. Edu. Technol. Soc. 13(3), 3–11 (2010)
36. Del Campo Cañizares, E.: M-Learning y aprendizaje informal en la educación superior
mediante dispositivos móviles. Ilu, vol. 18, pp. 231–242. https://doi.org/10.5209/rev_HICS.
2013.v18.44239
37. Nielsen, J.: Usability 101: Introduction to Usability. Nielsen Norman Group. https://doi.org/
10.1145/1268577.1268585. Articles
38. Schank, R.C., Fano, A., Bell, B., Jona, M.: The design of goal-based scenarios. J. Learn. Sci.
3(4), 305–345 (1994). https://doi.org/10.1207/s15327809jls0304_2
39. Vidal, M., Gavilondo, X., Rodríguez, A., Cuéllar, A.: Aprendizaje móvil. Revista Cubana de
Educacion Medica Superior 29(3), 669–679 (2015). https://doi.org/10.7238/rusc.v12i1.1944
40. Ally, M., Prieto, J.: What is the future of mobile learning in education? RUSC. Revista de
Universidad Y Sociedad Del Conocimiento 11(1), 142 (2014). https://doi.org/10.7238/rusc.
v11i1.2033
41. Paramo, P.: La investigacion en ciencias sociales: estrategias de investigacion, p. 332.
Universidad piloto de Colombia, Bogota (2013)
42. Lewin, K.: La investigación-acción y los problemas de las minorías. In: Salazar, M.C. (1992)
43. Carr, W., Kemmis, S.: Teoría crítica de la enseñanza. La investigación-acción en la
formación del profesorado. Martínez Roca, Barcelona (1988)
44. Lopez, L., Hernandez, J.: Estadística descriptiva, test y ejercicios, 2ª ed. Ediciones
Académicas. (EDIASA) edn, pp. 1–33. Madrid, España (2016)
45. Cofffey, A., Atkinson, P.: Encontrar el sentido a los datos cualitativos. Universidad de
Alicante, Alicante (2003)
46. Morgan, D.: Focus Groups as Qualitative Research. SAGE, Thousand Oaks (1997). https://
doi.org/10.4135/9781412984287
47. Onwuegbuzie, A., Teddlie, C.: A framework for analyzing data in mixed methods research.
In: Tashakkori, A., Teddlie, C. (eds.) Handbook of mixed methods in social and behavioral
research, pp. 351–383. SAGE, Thousand Oaks (2003)
Application of Cyberphysical Systems Through
Logistics as a Mediating Component
in Learning Processes
Andrés Felipe Córdoba U.(&), Clara Lucia Burbano Gonzalez(&),

Julio E. Mejía M.(&), and Jessica María Montilla M.(&)
Corporación Universitaria Unicomfacauca, Popayán-Cauca, Colombia

{andrescordoba,cburbano,jmejia,
jessicamontilla}@unicomfacauca.edu.co
Abstract. Technological advances make it possible to improve the industrial

sector, particularly in the management of production, one of the cases happens
in companies that use robots with decision-making capacity. This paper presents
the development of a serious game that integrates 3 logistic processes: organi-
zation, production, and transport, in addition to an evaluation process applied to
students of two academic programs of the Comfacauca University (Mechatronic
and Industrial Engineering), where 30 students are taken into groups of 15 for
each program. The data were subjected to the normality test using the Shapiro-
Wilk criterion and with the obtained results the T-Student parametric test was
applied which allowed establishing statistical differences between the groups. To
establish statistically significant differences P-Value less than 0,05 for produc-
tion and transport skills, while for the case of the organization no significant
differences were found.
Keywords: Education Industrial revolution Logistics Shapiro-Wilk

T-Student
1 Introduction
Logistics has been applied since the first era of human beings, being conceived as a
component in development, necessary for survival processes, a time when tribes were
in charge of storing, transporting and distributing raw materials or food. Later, in the
Middle Ages, the first technological advances were made: wheelbarrows, boats and
horse-drawn carriages, using them in processes applied to distribution, in order to
supply points of commerce in each community.
Then, the era of the industrial revolution arose, which integrated manual and
mechanical processes, leading to the manufacture of steam engines [1]. The industry
sought to improve the quality of life of human beings with the intention of optimizing
production processes. In the 1950s, the term “integral logistics” came into being,
defined by [2] as “the control of the flow of materials from the source of supply to place
the product at the point of sale, in accordance with customer requirements and with two
basic conditioning factors: a. Maximum speed in the flow of the product and b.

https://doi.org/10.1007/978-3-030-33229-7_20
Application of Cyberphysical Systems Through Logistics 231
minimum operational costs”, expecting companies to generate subcontracts with other

suppliers, thus dividing production processes at a lower cost, with high quality stan-
dards, satisfying the customer’s [3], the “Just in Time” method was originated,
inherently associated with technology, with the intention that industries improve
strategies linked to clients in logistics processes, obtaining as a result the elaboration of
a product in an exact time and quantity [4]. Since the year 2000, production processes
tend to be optimized on a large scale with complementary collaborative robotics in the
industry; it seeks to automate manufacturing lines using logistics as a main tool,
development plans aligned to manufacturing methods.
The companies reach the point of automating each process, fulfilling standards in
industry 4.0 [5]; in this way it is projected to industrialize the robotic stages for
production processes, defining a logistics plan in production stages: 1. Distribution of
goods in the company 2. Return of goods. 3. Product development 4. Waste man-
agement and 5. Selection of raw material; the processes have the possibility of being
executed by a robotic device in order to work collaboratively, grouping characteristics
of artificial intelligence, in business development, the case of manufacturing robots
with efficient tasks, capable of multiplying and overcoming human capabilities, among
them is articulated: a. Manage resources b. Distribute materials c. Transport mer-
chandise and d. Loading heavy elements [6]. Suffering an expansion: 1. Professional 2.
Work 3. Business and 4. Academic. The above refers to the importance of permeating
contents intervened in university careers in the Corporación Universitaria Comfacauca,
where students of Industrial Engineering and Mechatronic Engineering, from the sixth
semester incursion in the area of manufacturing processes, logistics and supply chain,
while students of Industrial Engineering continue their training in this field deepen in
topics such as distribution logistics and international trade. The above is intended to
infer how relevant it should be to integrate this type of process to train professionals
with technological skillsn [7], integrating current topics that can innovate the way they
do their work, in effect, “to know the innovations that lead to logistics 4.0, to develop
fluently in the digital environment, to master the new channels of management and
sales, such as eCommerce, and to be able to make decisions based on data are also
essential skills for the logistics professional of tomorrow” [8] and [9].
2 Methodology
The development of this research used a holistic approach, based on the methodology
proposed by Jacqueline Hurtado de Barrera, the project line is implemented, with seven
necessary phases. As it can be seen in Fig. 1, entitled “Research methodology”.
For each phase presented in the methodology a description was made: Phase 1:
Exploration based on antecedents, Phase 2: Description framed in technical and general
concepts; Phase 3: Describe the process, characteristics and aspects in Logistics; Phase
4: Diagnosis, recognize how the business sector is innovating; Phase 5: Description of
the technological resource; Phase 6: Reflection on the relationship of technology in
society and Phase 7: Conclusions.
232 A. F. Córdoba U. et al.
Fig. 1. Research methodology.
Additionally, a serious game evaluation was carried out in order to establish dif-
ferences between the results obtained from the two analysis groups; the first one of the
Mechatronic Engineering program (GEVIM) and the second one of Industrial Engi-
neering (GEVII), each one of the groups with 15 students who belong to the semester
of VI (Manufacturing Processes) and VIII (Logistics and Distribution) of the programs.
Each student was evaluated at each of the three levels through the score obtained and
the results were compared in order to establish the characteristics in relation to the
topics of 1. Organization Process 2. Production Process 3. Transportation process.
3 Conceptual Framework
One of the biggest challenges that faces the logistics industry today is the availability of
labor. It is not easy for companies around the world to find enough high-quality
employees to move products from customers to customers. Two conflicting factors
make this especially difficult: the first is the growing need for more logistics workers
and this is being driven by the e-commerce revolution and its need for more package
shipments; the second is a decrease in the size of the available workforce due to
declining population levels in the western world.
Robotics applied to logistics and collaborative work - robots - human or robots -
robots - performs work in predetermined spaces [6], fulfilling market strategies: 1.
Movement of resources 2. Support to staff 3. Inventories of materials and 4. Distri-
bution of goods [10].
The movement of resources implies increasing the effectiveness in relation to time,
this indicates that the robots that transport goods must have capacity to detect the
obstacles that represent a decision at the time of carrying out their work, these devices
have the capacity to geo-locate themselves by means of sensors and incorporated
cameras, to distribute goods or control the storage system, an example that currently bet
on this type of technologies are Amazon, Google, Vodafone, Fedex and Alibaba,
seeking to reduce their management of processes in less time.
In terms of storage of products in inventory, it is taken into account counting and

location within the store, each package has an identifier code in the registration system,
so the robot finds a package in the plant with efficiency of 20%, unlike a person; in
other words, a device lasts about 15 min, while an operator takes 90 min.
This evidences the contrast that exists between the optimization of processes by the
robot, the inventory systems currently use intelligent cameras that help to read bar
codes or QR with artificial vision, where they specify type of product and location in
the store, the robots have the ability to register each package in the software that
manages the company’s storage.
Platforms such as Kiva or Betty bot [11], have the ability to move heavy items to
operators in storage areas by facilitating freight forwarding processes. The distribution
of packages uses robots such as Wiki bot, FedEx, Sorthing Robot and livraison,
generating impact on society due to the autonomy of going to a point outside the
company delivering products: a. Fast foods b. Pharmaceutical drugs and c. Messaging
packages; this type of technology generates reliability when providing the service, so it
is necessary to use security codes or facial recognition of people.
3.1 Relationship Between Trend and Education

Education plays a fundamental role in carrying out research into the design and con-
struction of intelligent mechanisms applied in logistics processes, for which educa-
tional institutions update their programs focused on topics such as artificial intelligence,
artificial vision, neural networks, algorithm development and robotics.
The Corporación Universitaria Comfacauca offers programs oriented to a practical
learning in relation to Information Technologies mediated by laboratories with mbot
prototypes, necessary events for the logic of the collaborative work [12].
In addition to this, it can be used in logistics processes: 1. Implement robots to
organize and distribute elements in plants 2. Make use of robots in a collaborative way,
favoring group work 3. Promote Learning spaces in the field of Mobile Robotics
(SR) [13].
In the sense of understanding how learning is produced in the classroom [14], it
was necessary to appropriate different types of Logistics in the Training of Mechatronic
and Industrial Engineering professionals, generating a classification for the Logistics
processes, oriented to Functionality and Evolution.
By functionality of a company reflected in the organization in logistics, distributed
in 4 types as it can be illustrated in Table 1 entitled “Classification of logistics by
function”.
Table 1. Classification of logistics according to its function.

Types of Description
logistics
Procurement It is a set of tasks whose mission is to distribute and receive the
logistics proportional amount of raw material or goods requested
Distribution This type of logistics is aimed at the customer, which aims to transport the
logistics goods requested from it
Production For the development of the manufacture of the product, this logistics is
logistics presented, which is in charge of the elaboration and packing processes
Reverse The process of this logistics is very important, as it can occur in the case
logistics where the customer is dissatisfied with the purchase and wants to make a
return of the goods
By evolution in technology, evidencing an advance through time as shown in

Table 2 entitled “Classification of logistics according to its evolution”.
Table 2. Classification of logistics according to their evolution.

Logistics Description of technologies
versions
Logistics 1.0 In 1950, during the industrial revolution, the first steam engines were
introduced
Logistics 2.0 Then in the year 1970 the machines with greater distribution and
management of the products are illustrated with ports, shelves, and
platforms that work in an electrical way
Logistics 3.0 In the year 2000, industries began to use digital systems that were
responsible for managing the management of the plant, in addition to
designing a network that communicates the entire company
Logistics 4.0 For the present time the companies propose to make a global connection in
such a way that it is worked by subcontracting or as Smart Factory, mixing
the technology with the manpower for a better efficiency in the processes
The development of Logistics introduces us to understand the fundamentals in

regulations applied to the development of Logistics processes, as it is explained below:
3.2 Technical Standard in Robotic Safety

Collaborative robots work together with humans, performing repetitive work tasks,
most often on assembly lines or in CNC manufacturing functions. Unlike fixed robots,
which are programmed to operate independently of immediate human guidance, col-
laborative robots can work with human assistance because people can be in direct
contact with these types of robots, whether with guidance or monitoring, there are
stricter safety standards that apply to their operation. In this document, we will look at
the various types of operations and the standards that guide them.
In a company’s manufacturing processes, operators can cause an accident, in

relation to: 1. malfunction of the plant 2. damage to the mechanical or electrical part 3.
overloading of the robotic structure. Inadequate handling of equipment due to lack of
proper instruments, as indicated in the standard “ISO EN10218 Safety requirements for
industrial robots”. This standard is responsible for handling safety requirements in the
manufacture and integration of a robot into industry.
4 Technological Resource
In the development of the application it was decided to develop a serious game, in order
to appropriate knowledge, executing possible tasks that are carried out in a company.
For this purpose, the program Construct 2 was used as a development tool; creating
HTML5 applications that allow the implementation of 2D games, contributing to the
integration of learning processes for people who are not experts in the field of Logistics.
The game development process is shown in Fig. 2 named “Construct 2 Work Interface”.
Fig. 2. Construct 2 Work Interface.
The development of the game was focused in strategy mode with the tilemap
method, by means of different layers, where type layers is specified. The first layer
shows the main menu composed by the title of the game and three options, the second
option sets credits, and in addition to that you can select the exit button from the men.
The second layer constitutes the beginning of the game, demonstrating the main
objective aligned to interact within it, then press the accept button to continue with the
manipulation of the game. The game provides help information, explaining its intention
to execute a process at the shortest time, where each map corresponds to a logistics
process: 1. Organization Process 2. Production Process 3. Process of Transportation;
the fourth layer belongs to a menu of maps, evidencing three options that indicate each
logistic process, finally the option to return with the button return is offered. As it can
be seen in Table 3 named “ LogiBot Layers”.
Table 3. LogiBot Layers
Layer 1: Main Menu Layer 2: Aim of the Game
Layer 3: Help Layer 4: Layer Menu
The following layers comprise 3 generated maps, together with an evaluation

system, classifying the largest, taking as a variable the one that performs the work in
the least time and the valuation that proceeds. At the bottom of each map the score,
number of boxes, time, options to restart the game and the back button are presented: 1.
The first map conforms the organization process, which aims to group by type of
elements in 4 specific areas called A1, A2, A3 and A4, the main component lies in the
manipulation of the avatar conceived as a robot, changing its size depending on the
load it has. The other moving robots act as obstacles for the avatar, the second map is
production, the objective is to collect each of the elements: 1. Bottles 2. Tapes and 3.
Boxes, to then take them to the blue colored area, then the avatar enters the yellow zone
to activate a production mode, generating a box with the final product, taking it to the
store in the shortest time possible; the last map consists of a transport mechanism, the
idea is to collect a box of each color with the avatar: 1. Red 2. Green 3. White and 4.
Blue to go to the house that has the same color, it is there where the person receives it
and then moves to another box to meet the objective. This is presented in Table 4
named “LogiBot Maps”.
Table 4. LogiBot Maps
Map 1: Organization Map 2: Production
Map 3: Transportation
5 Discussion
The transformation in the industrial sector reflects changes from the way of managing
processes to the innovation of technological equipment used in production lines,
including cyberphysical systems; countries such as Japan attributed by its technological
scope, plans to develop robots on a large scale, wanting to fulfill tasks in any sector: 1.
Commercial 2. Industrial 3. Business and 4. Sports. One of the most important
advantages is the reduction of expenses at the moment of avoiding hiring employees; in
spite of advances obtained in technological field, allowing an efficiency and
improvement for the control of times, emphasizing that the robots have favorable
attributes, even better in performance, efficiency or force, making use of tools as the
artificial intelligence or autonomous robots in different contexts. The benefits lie in the
hyper-connectivity and automation of processes, properties highlighted under the rise
of the industrial revolution 4.0 that allows a better effectiveness at a productive level,
also it is evidenced by the transition from Digital User to Digital Citizen. In this context
a transformation takes place, this expands new forms of interaction, enabling intelligent
robotic systems, components such as: 1. Artificial Intelligence 2. Autonomous Vehicles
3. Internet of Things 4. Networked Systems 5. Big Data 6. Smart Cities 7. Drone
System and ultimately 8. Intelligent learning. The above is entitled cyberphysical
systems [14]. For the reader’s further illustration, a graph was used, showing the
contrast captured over time in the industrial revolutions from 1.0 to the current 4.0.
This is presented in Fig. 3 named “Evolution of the Industrial Revolutions”.
Fig. 3. Evolution of the industrial revolutions
In Education, refer to Intelligent Learning, authors such as Ros points out a type of
adaptive and contextualized learning with mechanisms of detection, response and
recommendation, supported by learning environments with technology, according to
the individual needs of students, where they practice analysis of learning behaviors,
performance trajectories, and online and real-world contexts. The possibility of inte-
grating artificial intelligence, making use of intelligent learning or Smart Learning in
robots applied to logistics, provides aid to learning, in relation to the fields of tech-
nology and robotics. Finally, it is concluded that social processes in relation to tech-
nology are associated with the training of Engineers in Fields of Mechatronics and
Industrial Engineering.
5.1 Relationship with Emerging Technologies

Since the 1800s the idea of recreating the first versions of robots with artificial intel-
ligence was generated, thanks to science fiction, which illustrates intelligent mecha-
nisms that make their own decisions, today this idea has not changed much to present
robots with neural networks that have the ability to perform high-risk work, by teaching
certain intervals of data to help it determine the right decisions. Currently the uni-
versities that develop programs in artificial intelligence identify 5 types which are
derived: 1. Voice 2. Vision 3. Robotics 4. Natural Learning Processing and 5. Machine
learning; however, the most appropriate way to apply an appropriate learning is sought,
supervised, not supervised or reinforced in such a way that towards a future it has the
capacity to think as if it were a human being when determining a decision with
Table 5. Trend Applications
1. Drones Spray Sequence 2. Drones in logistics for goods delivery
Taken Taken From: https://www.mediagroup.es/amazon-

from:https://www.linkedin.com/in drones-marketing-sevilla/
/jean-pierre-grimaud-77629696
3. Truck loading and unload-
4. Use of drones for terrain simulation
ing system
Taken from:
Taken from:
http://www.abconveyors.com.mx/
https://www.xdrones.es/levantamientos-
Soluciones-Roboticas/
topograficos-con-drones-fotogrametria-aerea/
5. Logistics in 3D printing
Taken from:
https://ultimaker.com/en/blog/52750-6-overlooked-benefits-of-3d-printing-for-your-
supply-chain
cognitive learning. On the other hand, this trend is related to knowledge of Big Data to
contribute to Data Learning that seeks to establish decisions with stored data to find the
right method. At present, educational projects are presented that take the characteristics
of students such as grades, stratum, income, class schedules, and that generate plat-
forms in academic processes.
5.2 Robotics Applied to Logistics

A variety of applications are presented: 1. Sprinkling sequence with drones: considered
as a high category project for business development, centralized in the irrigation of
agricultural crops, where its functionality allows to identify the type of plant selected,
ideal to generate a route within a geographical plane [15]. 2. Use of drones in logistics for
goods delivery. 3. Truck loading and unloading system: Companies such as Wynright
intend to implement intelligent robotic systems that contribute to optimizing logistical
processes for loading and unloading trucks [16], 4. Use of drones for land simulation:
This allows its application in the art of mining, providing a topographical recognition of
land with the support of an integrated vision system [17], 5. Logistics in 3D printing:
According to the contribution that [18]: the logistics applied in distribution of material
for several printers [19] and [20], in view of selecting type of material and exact quantity,
in order to perform specific tasks. This is shown in Table 5 named “Trend Applications”.
6 Results and Analysis
The students of each of the programs interacted with the game having only one
opportunity in each of the levels and the results obtained by them were recorded. It is
important to highlight that the score obtained by each of the participants takes into
Table 6. Results by maps - game use

User GEVIM GEVII
Map 1 Map 2 Map 3 Map 1 Map 2 Map 3
1 1235 1850 2050 1135 1420 1956
2 1482 1760 2350 1282 1610 2010
3 1356 1568 2200 1453 1386 1910
4 1236 1750 1968 1316 1530 1689
5 1357 1820 2010 1136 1220 2102
6 1206 1796 2120 1109 1264 2010
7 1420 1783 2169 1230 1573 2067
8 1230 1860 1965 1125 1402 1735
9 1502 1745 2200 1001 1552 2009
10 1750 1576 2189 1150 1368 1990
11 1330 1642 1985 1420 1523 1852
12 1386 1750 2010 1296 1602 1901
13 1278 1726 2120 1487 1662 1902
14 1310 1805 2150 1525 1452 1950
15 1420 1598 2210 1580 1184 1930
account the fulfillment of the objectives and the time spent in the development of each
map. In Table 6 named “results by maps - game use” you can see each of the ratings
obtained by the students.
The data were analyzed descriptively in order to understand the behavior of the data
obtained for each of the analysis groups, then in Table 7 named “descriptive statistical
results for collected data”.
Table 7. Descriptive statistical results for collected data

Group Mean Standard deviation
GEVIM Map 1 1,366.53 140.44
GEVIM Map 2 1,734.66 96.87
GEVIM Map 3 2,113.07 111.99
GEVII Map 1 1,283.00 176.49
GEVII Map 2 1,449.87 147,16
GEVII Map 3 1,934.20 112.13
As it can be seen from the Table, the mean values for map 1 do not differ sub-
stantially. The opposite occurs in relation to maps 2 and 3 of each of the groups where
the calculated mean values have differences of 284.79 and 178.87 correspondingly in
favor of the results obtained by the Industrial Engineering analysis group. In relation to
the deviation of the data a very similar behavior is presented showing uniformity in the
groups of evaluation that can be seen in the Fig. 4 denominated “Radial graph for
abilities of the groups of analysis”.
Fig. 4. Radial graph for abilities of the groups of analysis

Table 8. Shapiro-Wilk test for normality analysis of the knowledge test performed. Source own
elaboration
Group Statistics gl * P-Value
GEVIM1 0.944 15 0.438
GEVIM2 0.948 15 0.494
GEVIM3 0.932 5 0.293
GEIIM1 0.881 5 0,051
GEIIM1 0.893 15 0.073
GEIIM1 0.928 15 0.253
As can be seen in Fig. 4, the results obtained present a greater difference for the
concepts of production, this is mainly because the students of Mechatronics Engi-
neering do not adopt this concept within their academic curriculum. The data collected
were subjected to normality tests to subsequently perform inferential analysis of the
data in Table 8 called “Shapiro-Wilk test for normality analysis of the knowledge test
performed.” You can see the values obtained for the test statistic, for this case we
proceeded to perform the “Shapiro-Wilk test” [21] highlighted by its frequent use in
case studies and with better validity at the statistical level, it is applied in groups with
samples less than 50 individuals. The normality test raises a null hypothesis (Ho) re-
lated to the origin of the data from a normal distribution and an alternative hypothesis
(Ha) that states, as the data deviates significantly from a normal distribution; To val-
idate the hypothesis, the bilateral asymptotic significance of the test (p-Value) is taken
into account.
As can be seen in the values obtained for each group analyzed, the P-Value
obtained is greater than 0.05 (P > 0.05) which allows to accept the null hypothesis
(Ho) and affirm that the data collected comes from a Normal distribution, these results
obtained from the normality analysis allowed us to apply a parametric analysis of the
data through the T-Student test.
To analyze the differences between the groups, the T-Student test was applied. The
t-test directed to independent samples that allow comparing means of two groups,
indicating if there is statistical evidence where the associated population means are
significantly different using bilateral asymptotic significance. (P-Value). For this case it
was defined as null hypothesis (Ho) the non-existence of differences between the means
of the analytical skills and alternative hypothesis (Ha) the existence of difference
between the means of the analysis skills, They were made to the different skills, in
order to verify the existence of differences between the results obtained for each of the
maps related to the logistics processes.
Table 9. T-Student test for analysis of difference of means. Source own elaboration
T-test for equality of means
Ability t gl Sig. Mean Typ. error of the
(bilateral) difference difference
Organization 1,434 28 0.163 83.53 58.23
Production 6,261 28 0,000 284.8 45.49
Transport 4,371 28 0,000 178.8 40.91
As it can be seen in the Table 9, the value of bilateral asymptotic significance is less
than 0.05 (P < 0.05) for the Production and Transport cases, which leads to rejecting
the null hypothesis Ho and accepting the alternative hypothesis Ha, allowing to
establish that the scores obtained by the different groups generate differences depending
on the academic program. In the case of organizational ability, it can be observed that
the bilateral asymptotic significance is greater than 0.05 (P > 0.05), which allows
accepting the null hypothesis Ho, establishing that this competition in the organiza-
tional field does not present a difference between the two groups.
7 Conclusions
The adaptation of technological tools when applying a serious game allows to establish
didactic methodologies to evaluate the technical and technological competencies
related to logistics processes, the results obtained from the application of the t-Student
test allowed to establish statistically significant differences obtaining a P-value less than
0,05 (P < 0,05), allowing to conclude that the Mechatronics and Industrial Engineering
groups presented significant differences related to Logistics and transportation skills, in
relation to the organizational skill, did not appear statistical.
The use of a serious game related to robotics applied to logistics processes signifi-
cantly improved the Learning processes within the subject of manufacturing and logistics
processes in the Mechatronics and Industrial Engineering program of the Comfacauca
University on the other hand. The combination of technological and communication
tools Smart Factory intends to maintain an appropriate solution for efficiency, punctu-
ality, productivity and quality in the services of the company, allowing to favor mean-
ingful and collaborative learning in professionals that enhanced their logistics.
The use of Robotics, in industrial environments, in Revolution 4.0, integrates human
capabilities with potentials of robots such as repetitiveness, speed, strength, security,
allowing to improve logistics processes, about to efficiency and science. efficiency
increasing up to 20% in pre-established work in the activities developed by the operators.
References
1. Barros, T.: La Industria 4.0: Aplicaciones e Implicaciones, Universidad de Sevilla, pp. 1–52
(2017). http://bibing.us.es/proyectos/abreproy/91146/fichero/La+Industria+4.0+Aplicaciones
+e+Implicaciones.pdf
2. Anaya, J.: Logística integral: La gestión operativa de la empresa, pp. 25–27. Libros
profesionales. ESIC Editorial, Madrid (2015)
3. Sarria, I., Lozano, B., Tenorio, W.: MÉTODO JAT JUSTO A TIEMPO, pp. 1–96.
Universidad Libre Seccional, Cali (2015)
4. Tipping, A., Prümm, D., Kauschke, P., Smith, J.: Shifting patterns. PWC 11 (2016). http://
www.ncbi.nlm.nih.gov/pubmed/8974216
5. Azmi, I., Hamid, N.A., Nasarudin, M., Hussin, M., Ibtishamiah, N.: Logistics and supply
chain management: the importance of integration for business processes. J. Emerging Econ.
Islamic Res. 5(4), 73–80 (2017). http://www.jeeir.com
6. Sanz, J.: Robots industriales colaborativos: Una nueva forma de trabajo. Segurdad y salud en
el trabajo 95, 6–10 (2018)
7. Bensusán, G.: Los mecanismos de fijación de los salarios mínimos en México en una
perspectiva comparativa: el marco institucional y los actores sociales. In: Mancera, M.A.
(Coordinador), Del salario mínimo al salario digno, pp. 205–253. Consejo Económico y
Social de la Ciudad de México, México (2015)
8. Roig, M.V.: Logística y cadena de suministro en la nueva era digital. Revista de economía,
empresa y sociedad (2018)
9. Chaparro-Peláez, J., Agudo-Peregrin, A.F., Pascual-Miguel, F.J.: Conjoint analysis of
drivers and inhibitors of E-commerce adoption. J. Bus. Res. 69(4), 1277–1282 (2016)
10. Rayo, J.: La logística en el sector industrial. Logistics Assistant en Técnicas Reunidas 12,
64–69 (2016)
11. Wu, Y., Ge, D.: Key technologies of warehousing robot for intelligent logistics. In: The First
International Symposium on Management and Social Sciences (ISMSS 2019). Atlantis Press
(2019)
12. Patiño, J.G., et al.: La cuarta revolución industrial. Ingenierías USBMed 10(1), 1 (2019)
13. Báez Sánchez, X.E.: Efectividad del “Robot Milo” en el desarrollo de habilidades sociales y
comunicación en niños de 5 a 7 años con trastorno del espectro del autismo de grado 1.
Bachelor’s thesis, Quito (2018)
14. Barrera, P.: Diseño de un en el área de la logística organizacional sistémico proyecto
educativo mediante las tic. Universidad Militar Nueva Granada 7(2), 1–21 (2017)
15. Amortegui, D., Betancurt, J., Soler, A.: Diseño de un sistema autonomo para fumigacion
aerea (2014)
16. Mendez, J.: Logística colaborativa como herramienta para mejorar el nivel de servicio y
disminuir los costos de distribución. Universidad militar nueva Granada, pp. 1–20 (2018).
https://repository.unimilitar.edu.co/bitstream/handle/10654/17745/
MendezBohorquezJulioAlexander2018.pdf?sequence=1&isAllowed=y
17. García, M.S.: Uso y aplicaciones de drones en minería (2017). http://bio-digestores.blogspot.
com.co/2012/06/uso-y-aplicaciones.html
18. Lopez, J.: Fabricación aditiva y transformación logística: la impresión 3D. Oikonomics 9,
58–69 (2018). http://oikonomics.uoc.edu
19. Telefónica, F.: Sociedad Digital en España 2018. Fundación Telefónica (2019)
20. del Val Román, J.L.: La Transformación Digital de la Industria Española. In: Conferencia de
Directores y Decanos de Ingeniería Informática Ind, p. 10 (2016). https://doi.org/10.1080/
14015430802688385
21. Dean, A., Morris, M., Stufken, J., Bingham, D.: Handbook of Design and Analysis of
Experiments, 1st edn, pp. 140–148. Chapman and Hall/CRC, Boco Raton (2015)
Internet of Things
IoT Botnets
Pamela Beltrán-García, Eleazar Aguirre-Anaya(&),

Ponciano Jorge Escamilla-Ambrosio, and Raúl Acosta-Bermejo
Laboratory of Cybersecurity, Centro de Investigación en Computación,

pam.belt.g@gmail.com,
{eaguirre,pescamilla}@cic.ipn.mx, racostab@ipn.mx,
http://www.cic.ipn.mx
Abstract. This paper presents a comprehensive state-of-art review that dis-

cusses the IoT botnet behaviour, including topology and communication
between botmaster and bots, thus is possible to make a comparison of IoT
botnets, based on their topology, type of attack, target, kind of propagation and
operation. In several investigations, it is explained that a significant problem is
an increase in the development of IoT botnets, such as attacks like DDoS. To
this aim, understanding the behaviour of the IoT botnets could be helpful to
prevent them.
Keywords: IoT Botnet Bot
1 Introduction
The concept of Internet of Things was initially expressed as “computers everywhere,”

formulated by Ken Sakamura at the University of Tokyo in 1984. In 1999, Kevin
Ashton was the first to devise the term “Internet of Things” [1]. The phrase “Internet of
Things,” which is also shortly well-known as IoT composed of two words: first is
“Internet” and second is “Things.”
The Internet is a global system of interconnected networks that use the standard
Internet protocol suite (TCP/IP) to serve billions of users worldwide [3]. Devices with
few resources in computation and energy capacity characterize the term “Things” in
IoT; these can have sensors, actuators, and processing unit; which have an IP address
assigned for Internet connectivity, either wired or wireless networks [2, 4]. Due to the
few IoT devices resources, complicates the task of installing security controls and
causes them to be vulnerable to be infected and execute DDoS attacks by flooding a
service with legitimate requests [27].
With the increase in the development of IoT devices, security has become an
essential factor due to attacks that are multiplying; one way is the use of botnets. Silva
et al. [24] discuss that approximately 16–25% computers connected to the Internet are
members of botnets. Other studies report that at the beginning of the year 2008, e-mail
spam was generated only by six botnets [25]. Also, Symantec Internet Security Threat
Report indicates that 5.06 million distinct botnet computers; 61,940 active computers
per day, and 4,091 bot command servers have been observed [26].

https://doi.org/10.1007/978-3-030-33229-7_21
248 P. Beltrán-García et al.
Actually, according to Spamhaus Malware Labs, 10,263 botnets hosted on 1,121

different networks in 2018 were identified and blocked. That is an 8% increase from the
number of botnet C&C in 2017 [32].
2 Botnets Background
The botnet term is a combination of two main words, Bot as a Robot abbreviation and
Net for Network; so, it can be defined as a “network of infected hosts called bots, which
are controlled by a human operator, better known as the Botmaster.”
Botnets recruit vulnerable hosts by using a type of malware that exploits vulner-
abilities; an infrastructure is created between the now infected hosts and the Botmaster
who takes the control of these hosts remotely using a Command and Control server
(C&C), through the sending of commands to perform malicious activities, such as
DDoS, spam, or information theft.
In the context of IoT, botnet is a network of compromised IoT devices, such as
cameras, modems, DVR, sensors and other devices that use the IP protocol with the
characteristic to transmit data over the Internet, infected with malware. Such networks
allow an attacker to control the devices performing malicious tasks, as well as prop-
agate their malware [5].
2.1 Botnet Stages

Botnets perform their actions in three main stages [7, 8]:
• Infection: The malware used for recruit new bots can be propagated by exploiting
vulnerabilities, downloading by web/mail, installing software from unreliable
sources, then executed, and the host became part of the botnet.
• Command and Control: Once infected, the bot communicates with the host that
controls the botnet to receive commands.
• Malicious Activities: Execution of attacks such as DDoS, spam, etc.
2.2 Botnet Topologies

According to Vormayr et al. [9] and Dhinnesh et al. [10], the topology of the botnets is
defined by the command and control process they execute:
• Centralized: The Botmaster controls and monitors all bots from a single central
point, which makes the latency low, that is, all bots receive commands and reports
to the center point (C&C server). Likewise, there are two centralized topologies:
Star and hierarchical, in which the Protocols Internet Relay Chat (IRC) and Hyper
Text Transfer Protocol (HTTP) are mostly used; like in the Chuck Norris or Aidra
IoT botnet (Fig. 1(a)).
• Decentralized: Also known as Peer-to-Peer (P2P), in which the bot acts as a server
and client, each one is connected to another bot at least. The commands can reach
each bot only if all the bots are connected (Fig. 1(b)).
IoT Botnets 249
Fig. 1. (a) Centralized [9], (b) Decentralized [9] and Hybrid [9]
• Hybrid: It´s a combination of centralized and decentralized topologies since there

are two types of bots, some of them have the functionality as servers and clients
others just as clients, makes the message latency high (Fig. 1(c)).
Authors in [9, 11, 12, 28], described the communication between the C&C server
and bots, as shown in Table 1.
Table 1. Communication C&C server and bots

Advantages Disadvantages IoT Botnet application
IRC Channel Use of a centralized topology, Close ports that were used
protected by this means a single failure point to gain access to IoT
authentication for the botnet device, i.e., like in the IoT
mechanisms Linux/Hydra botnet
HTTP Use of a centralized or P2P Such as the Mirai botnet
topology, in which case has to gained remote access to
prevent loops or replicated the device over Telnet,
messages SSH, or HTTP
SMB Often blocked by a gateway, so Bashlite IoT Botnet makes
is mainly used for local use of this channel by
networks downloading two scripts
to gain remote access,
then inject malicious code
into files, generating keep-
alive messages exchanged
every 60 s between the
botnet C&C and the bot
As mentioned in Sect. 2.1, for an IoT botnet the main botnet stages are the com-
munication between the botmaster and the bots due to the botmaster does not have
direct communication with the bots, it makes use of the C&C server.
According to the topology and communication channel, botnet detection can be

attached by building a C&C server detector.
2.3 Botnet Communication

Therefore, an exchange of messages sequence is needed to achieve a specific task:
• Coordination: If the particular task is not automatic, it is necessary to instruct the
bot what to do.
• Scan: Network scan, can be an ICMP echo request, UDP or TCP scan for vul-
nerable services.
• Data: It can be a binary bot, arbitrary files, or network data that communicates to the
botnet or other bots.
• Register: Message required for a record.
• Execute: Tasks executed by bots.
As well, it can be separated as presented in Table 2.
Table 2. Botnet stages

Infection Operation
• It is used to recruit new bots • The C&C sends commands
• It can be active (when the botnet tries to infect additional to the active bots
hosts) or passive (when the binary bot is distributed by • The stage in which botnet
other means) performs the task
According to Vormayr et al. [9] botnets continuously recruit new bots by infecting
hosts for launching a specific attack, this infection stage occurs in two different ways as
exposed in Table 3.
Table 3. Infection stage

Active infection Passive infection
• Makes use of existing vulnerabilities • Via email, web pages, or storage media
• Through commands or automatically • Users infect their host by a click or an
1. Coordination: Scan and parameters action
configuration to be exploited • The binary bot is executed on the infected
2. Scanning: Host detection achievable with an host
ICMP echo request or directly in the • Downloading additional data
vulnerabilities of the UDP and TCP ports 1. Register: Monitor the botnet status, as
3. Infection: Vulnerabilities exploitation well as the size and location of the bots
4. Download additional data
5. Registration: Monitoring the botnet status, as
well as the size and location of the bots
IoT Botnets 251
These two infection ways are applied to IoT devices due to, i.e., infected cameras
by exploiting existing vulnerabilities such as default credentials; or like tablets where
the user infects their device by a click allowing the binary bot download, these being
helpful to understand botnet communication and therefore install detection
mechanisms.
2.4 Communication Hiding and Obfuscation

To increase the probability of survival, botnets tries to evade detection using hiding and
obfuscation techniques in the operation stage [9, 12]:
• Covert channels: They are used to hide the existence of communication.
• Encryption: Used to hide transmitted commands as well as the protocol used. If a
botnet uses a predefined key and fixed commands, the botnet can be disabled.
• Compression: A compression algorithm can hide the data.
• Steganography: Hiding information in non-communicating containers, similar to
covert channels with the difference of being used as carriers.
Knowing botnet communication and obfuscation techniques, network-based sig-
nature detection can be applied. According to Vormayr et al. [9], the operation stage
can be categorized, as shown in Table 4.
Table 4. Operation stage

Operation stage Actions
Data Upload: Data collected from bots • C&C sends commands to bots with specific
instructions
• Instructions are executed
• Report results to C & C
Data Download: storing data onto bots, • C&C indicates the bot which file to download
i.e., botnet binary update and where to find it
• Download file in the bot
• Store or install the download
Forward Proxy: Uses the botnet to hide • Unidirectional: It consists of data sent from the
the actual origin of the communication bot to a target, used to overload a single
service with requests
• Bidirectional: It consists of requests and
responses that transmit through the bot, used if
the Botmaster or a third entity needs the results
of a request
Reverse proxy: Reverse for • C&C indicates the bot which port to open and
retransmission to a specific source host which internal address should connect so that
other hosts or bots can connect to the source
Instruction: Execute tasks on behalf of • The botmaster indicates the bots what tasks
the Botmaster must be executed and provides the necessary
parameters
• The current command executes
During the operation stage, the sequence of exchanging messages depends on the
botnet specific task, IoT botnets examples are in Sect. 3.
Studying the exchanging messages could be possible to collect information packets
and determine the type of botnet used so that detection mechanisms can be used.
3 IoT Botnet Attacks
The use of a botnet is for several purposes such as malicious activities, better known as
attacks. In a DDoS attack, the network bandwidth is consumed by the compromised
IoT devices injecting malicious packets into the network targeting a particular server;
this means it floods the traffic with service request and processed by a server [29, 30].
Specifically, the most executed IoT botnets flooding attacks are:
• TCP SYN Flood: exploits a known weakness in the TCP connection sequence, the
requester sends multiple SYN requests, but either does not respond to the host’s
SYN-ACK response or sends the SYN requests from a spoofed IP address. Either
way, the host system continues to wait for acknowledgment for each of the requests,
making binding resources until no new connections, and ultimately resulting in a
denial of service.
• UDP Flood: a large number of User Datagram Protocol (UDP) packets are sent to a
targeted server to overwhelm that device’s ability to process and respond.
• HTTP Flood: the attacker exploits seemingly-legitimate HTTP GET or POST
requests to attack a web server or application.
• ICMP Flood: overwhelms the target resource with ICMP Echo Request (ping)
packets, generally sending packets as fast as possible without waiting for replies.
• ACK/ACK PUSH Flood: receives no legitimate ACK packets that do not belong
to any of the sessions on the server’s list of transmissions. The server under attack
then wastes all its system resources (RAM, processor, etc.) trying to define where
the packets belong. The results in productivity loss and partial server unavailability.
• TCP XMAS Flood: Send a very explicitly crafted TCP packet to a device on the
network with FIN, URG and PUSH flags set.
• DNS Flood: Attackers send validly but spoofed DNS request packets at a very high
packet rate and from a large group of source IP addresses. The DNS server can be
overwhelmed by the number of requests.
A single IoT device as bot is not a threat, but the recruitment of several bots can
attached flooding attacks, due to these devices can send connections requests affecting
the availability of services. By determining the type of a botnet attack, it is possible to
apply a network-anomaly detection technique.
4 Related Work
As briefly mentioned in Sect. 3, IoT botnets are performing many attacks. In this
section, an IoT botnet comparison is done at Table 5 based on model architecture,
attacks, target, operation, and propagation [6, 13–23, 28].
IoT Botnets 253
Table 5. IoT botnets

Botnet Archit. Attacks Target Infection Operation
model
Linux/Hydra (2008) IRC SYN & UDP Routing Active: Download data:
Flood devices based Dictionary the attacker had
on MIPS attack, known to edit one of
architecture specific the source files
authentication to provide the
vulnerability URL address of
the C&C IRC
server as well
as the link to
download the
malicious
binary
Psyb0t (2009) IRC SYN, UDP & Routers & Active: Access Download
ICMP Flood DSL modems to Telnet and Data: Once a
SSH using shell of the
brute force with vulnerable
6000 predefined device is
user names and acquired,
13000 Psyb0t
passwords downloads
itself from a
remote server
Chuck Norris (2010) IRC SYN, ICMP & D-Link Active: Brute Download
ACK Flood routers & force, as well as Data:
DSL modems the Download SSH
authentication
override
vulnerability
Tsunami (2010) IRC SYN Flood, Linux Mint Passive: Download
UDP Flood & Official ISO Modifies the Data:
ACK DNS server Download file
Also, HTTP setting in the (s) from remote
Flood & TCP configuration of servers
XMAS the infected
devices such
that the traffic
redirects from
the IoT device
to malicious
servers
LightAidra/Aidra IRC SYN Flood & Architectures Active: Search Download
(2012) ACK Flood such as open Telnet Data:
MIPS, ARM, port using Download a
and PPC default shell script from
credentials buonapesca.
Propagates over altervista.org
TCP on port 23, The script
for install a downloads and
backdoor. executes
(continued)
Table 5. (continued)
model
additional
malicious files,
also receive a
command from
the IRC server to
perform the
attack
The Moon (2014) P2P Bring down Linksys, Active: Reverse proxy:
websites and ASUS, Exploiting a download an
servers by MikroTik, command additional
overwhelming and D-Link vulnerability in proxy module
them with routers the POST that opens a
requests; request SOCKS5 proxy
obfuscate parameter on infected
information devices, who
online knows the web
page and the
parameters to
which the
redirection of
the uplink node
must be
accessed, and
the port no
longer opens
directly on the
infected node
Bashlite (2014) IRC SYN, UDP, Linux-based Active: Brute Upload Data: In
HTTP & ACK IoT devices force with the malware´s
Flood such as default binary has the
cameras and credential of IP address of
DVRs devices with the C&C server.
open Telnet It also has the
ports IP addresses
hard-coded into
it
Mirai (2016) Centralized Generates Closed-circuit Active: 10 Upload Data:
floods of television, predefined loading the
GRE IP, cameras, attack vectors malware to the
GRE ETH, routers, and Dictionary vulnerable IoT
SYN, ACK, DVR attack based on devices
STOMP, DNS, 62 entries detected, scan
UDP, & HTTP the network for
new victims
while waiting
for instructions
from the C&C.
Scan traffic uses
random
(continued)
IoT Botnets 255
Table 5. (continued)
model
parameters to
avoid
identification
and
fingerprinting
Linux/IRCTelnet/New IRC SYN, ACK- Routers, Active: Brute Upload Data: In
Aidra (2016) PUSH, & UDP DVR, and IP force and code the malware´s
Flood cameras injection, list of binary has the
Mirai IP address of
credentials the C&C server.
It also has the
IP addresses
hard-coded into
it
Table 5 lists the botnets analyzed in this work along with their topology, targets,
types of attacks and the infection/operation stages studied in Sect. 2; each column gives
information about how the IoT botnet operates, making possible the use of detection
mechanisms. Some mechanisms such as signatures-bases detection are referring to the
analysis of known or abnormal patterns or characteristics of threats from intruders into
a system, or by honeypots, which are used as traps to collect bot´s information and
activities, making possible to analyze them to detect botnets. These detections are
performed when the attack is executed [31].
5 Discussion
Botnets grow in size and complexity as potential nodes in the Internet of Things time to
time, making difficult to identify malicious from benign traffic even monitoring the
common ports used, like 23,2323 and 22 to gain access to the IoT device.
From a diverse set of IoT attacks described in Sect. 3 and the botnet communi-
cation in Sect. 2, different botnets were chosen in the state-of-art to identify the type of
infection and operation used. These botnets have been analysed according to their
topology, target, and type of attack, for determining the way it infects new hosts and
how it operates.
It was observed similar characteristics in most IoT botnets, such as centralized
topology due to the IoT devices resources cannot function as a server the main cons of
this topology is the single failure point by detecting the C&C server, although the pros
are management and monitoring of the botnet by the botmaster that communicates
directly with each bot, and the low latency.
As compare to decentralized and hybrid topology, there are more than one C&C
server, and because of the IoT devices resources, these can not function as a server,
these are cons. Although, the pros lies in the detection complexity of a failure point
because if one C&C server is taking down, the other servers can manage and monitor
the botnet.
The application of a centralized topology still exists, by creating new botnets
following the evolution of applications on the Internet. According to the previous
research, it is provided a comprehensive state-of-art review of botnets that are evolving
time to time making important and understandable point their application to a new
target, IoT devices, highlighting how evolution and behaviour can be expanded to
perform different attacks in the IoT context.
6 Conclusion
This paper presented a comprehensive review and analysis that discusses the IoT botnet
development with distinct variation targets, attacks, and type of propagation and
operation. Being helpful to have a better knowledge of the communication between
different IoT botnets that use similar techniques or a combination of them, so that
botnet prevention can be created or even disable the botnet, i.e., the use machine
learning to extract message exchanges from network traffic for identifying possible
botnet communication.
Acknowledgment. The authors would like to thank the Instituto Politécnico Nacional (IPN), the
Centro de Investigación en Computación (CIC) and the Consejo Nacional de Ciencia y Tec-
nología (CONACYT) for the support in this research.
References
1. Escamilla-Ambrosio, P.J., Rodríguez-Mota, A., Aguirre-Anaya, E., Acosta-Bermejo, R.,
Salinas-Rosales, M.: Distributing computing in the internet of things: cloud, fog and edge
computing overview. Stud. Comput. Intell. 731, 87–115 (2018)
2. Madakam, S., Ramaswamy, R., Tripathi, S.: Internet of Things (IoT): a literature review.
J. Comput. Commun. 03(05), 164–173 (2015)
3. Nunberg, G.: The Advent of the Internet: 12th April, Courses (2012)
4. Stavrou, A., Voas, J., Fellow, I.: DDoS in the IoT Mirai and Other Botnets-2017-Computer
(2017)
5. Botnet de IoT Homepage (botnet de internet de las cosas). https://searchdatacenter.
techtarget.com/es/definicion/IoT-botnet-botnet-de-internet-de-las-cosas
6. Angrishi, K.: Turning Internet of Things (IoT) into Internet of Vulnerabilities (IoV): IoT
Botnets, pp. 1–17 (2017)
7. Tyagi, A., Aghila, G.: A wide scale survey on botnet. Int. J. Comput. Appl. 34(9), 9–22
(2011)
8. Zhaosheng, Z., Zhi, J.F., Guohan, L., Phil, R., Yan, C., Keesook, H.: Botnet research survey.
In: Proceedings of the International Computer Software and Applications Conference,
pp. 967–972 (2008)
9. Vormayr, G., Zseby, T., Fabini, J.: Botnet communication patterns. IEEE Commun. Surv.
Tutorials 19(4), 2768–2796 (2017)
10. Sundareswaran, N.: Botnet life cycle and topologies. Int. J. Pure Appl. Math. 119(17), 421–
429 (2018)
IoT Botnets 257
11. Dwivedi, S.K., Bist, A.S., Chaturvedi, P.K.: Recent trends in botnet research. Int. J. Eng.
Sci. Res. Technol. 6(7), 280–295 (2017)
12. Khattak, S., Ramay, N.R., Khan, K.R., Syed, A.A., Khayam, S.A.: A taxonomy of botnet
behavior, detection, and defense. IEEE Commun. Surv. Tutorials 16(2), 898–924 (2014)
13. De Donno, M., Dragoni, N., Giaretta, A., Spognardi, A.: DDoS-capable IoT malwares:
comparative analysis and mirai investigation. Secur. Commun. Netw. 2018, 1–30 (2018)
14. Spognardi, A., De Donno, M., Dragoni, N., Giaretta, A.: Analysis of DDoS-capable IoT
malwares. In: Proceedings of the 2017 Federated Conference on Computer Science and
Information Systems, vol. 11, pp. 807–816, September 2017
15. Meidan, Y., et al.: N-BaIoT-network-based detection of IoT botnet attacks using deep
autoencoders. IEEE Pervasive Comput. 17(3), 12–22 (2018)
16. Durfina, L., Kroustek, J., Zemek, P.: PsybOt malware: a step-by-step decompilation case
study. In: Proceedings of the Working Conference on Reverse Engineering WCRE, pp. 449–
456 (2013)
17. Hallman, R., Bryan, J., Palavicini, G., Divita, J., Romero-Mariona, J.: IoDDoS – the internet
of distributed denial of service attacks. In: IoTBDS (2017)
18. Janus, M.: Heads of the Hydra. Malware for Network Devices, 16 August 2011. https://
securelist.com/heads-of-the-hydra-malware-for-network-devices/36396/
19. Barnett, R.: New Tsunami/Kaiten Variant: Propagation Status, 11 September 2018. https://
blogs.akamai.com/sitr/2018/09/new-tsunamikaiten-variant-propagation-status.html
20. Symantec “Linux.Aidra” Writeup By: Kaoru Hayashi. https://www.symantec.com/security-
center/writeup/2013-121118-5758-99
21. Cyware “The Moon IoT botnet is proxying traffic for Youtube ad fraud scheme”, 1 February
2019. https://cyware.com/news/themoon-iot-botnet-is-proxying-traffic-for-youtube-ad-fraud-
scheme-e17d6945
22. Netlab website. https://blog.netlab.360.com/themoon-botnet-a-review-and-new-features/
23. NJ Cybersecurity and Communications Integration Cell “Linux/IRCTelnet”, 3 November
2016. https://www.cyber.nj.gov/threat-profiles/botnet-variants/linux-irctelnet
24. Silva, S.S.C., Silva, R.M.P., Pinto, R.C.G., Salles, R.M.: Botnets: a survey. Comput. Netw.
57(2), 378–403 (2013)
25. AsSadhan, B., Moura, J.M.F., Lapsley, D., Jones, C., Strayer, W.T.: Detecting botnets using
command and control traffic. In: Proceedings of the 2009 8th IEEE International Symposium
on Network Computing and Applications NCA 2009, no. 4, pp. 156–162 (2009)
26. Symantec Internet Security Threat Report: Trends for July–December 2007 (Executive
Summary), vol. XIII, April 2008
27. Makhdoom, I., Abolhasan, M., Lipman, J., Liu, R.P., Ni, W.: Anatomy of threats to the
internet of things. IEEE Commun. Surv. Tutorials 21(2), 1636–1675 (2019)
28. Ceron, J.M., Steding-Jessen, K., Hoepers, C., Granville, L.Z., Margi, C.B.: Improving IoT
botnet investigation using an adaptive network layer. Sensors 19(3), 1–16 (2019)
29. Ahmed, M.E., Kim, H.: DDoS attack mitigation in internet of things using software-defined
networking. In: Proceedings of the 3rd IEEE International Conference on Big Data
Computing Service and Applications, BigDataService 2017, pp. 271–276 (2017)
30. Gupta, B.B., Badve, O.P.: Taxonomy of DoS and DDoS attacks and desirable defense
mechanism in a cloud computing environment. Neural Comput. Appl. 28(12), 3655–3682
(2017)
31. Lange, T., Kettani, H.: On security threats of botnets to cyber systems. In: 6th International
Conference on Signal Processing and Integrated Networks, SPIN 2019, pp. 176–183 (2019)
32. Spamhaus Malware Labs: Spamhaus Botnet Threat Report 2019, pp. 1–15 (2018)
Traffic Light Control Through Intelligent
Agents, Sensors and Arduino
René Alberto Ojeda-Cepeda, Larissa Jeanette Peniche-Ruiz(&),

Iván Antonio Castro-Ynurreta, and Edwin Moises Pool-Couoh
Tecnológico Nacional de México, Instituto Tecnológico de Mérida,

Mérida, Yucatán, Mexico
rene_ojeda_96@hotmail.com, larissa.peniche@itmerida.mx,
castroynurreta@gmail.com, edwinpoolcouoh@gmail.com
Abstract. Car traffic control is an everyday problem in a city large enough to

hold more than a million population. This paper presents an intelligent traffic
light control system based on intelligent agents, Arduino and sensors, which
produced data is used as entry to a scale model of a busy street intersection. The
project succeeds at managing car flow on the model street crossing allowing to
set variables such as time of the day, emergency vehicles and number of cars on
each street to display solutions to traffic jam situations.
Keywords: Intelligent agents Traffic control Sensors
1 Introduction
Traffic congestion in the cities of the world is a problem which directly impacts both in
the journey time and in the number of car accidents. This kind of incidents increase
during rush hour with well-known consequences like delays, drivers and passengers
annoyance, material damages and even casualties. Traffic jam consequences often end
up in car accidents mostly because drivers lose control when they are forced to wait for
a long time in a saturated street, which in turn stresses drivers to the point of speeding
up when the street is clear. The excess of traffic may evolve into violent episodes,
disturbing other drivers and increasing time loss and traffic congestion. In the same way
other collateral damages may be produced, like excessive gas consumption while cars
wait in line for a long time in the same place, thus environmental pollution with CO2
rises considerably [1–3].
The most common system in the cities to control car flow is the traffic lights, which
basically work with green, yellow and red lights to signal go, go with caution and stop,
respectively. Their aim is to regulate and speed up car flow in the streets. However,
traffic lights nowadays do not have the ability to intelligently manage traffic congestion
on a public street at any given time; traffic lights work with timing devices which allow
cars flowing on streets according to synchronized allotted time slots. There are a few
traffic lights which consider the general state of the surrounding driveways to adjust
their performance intelligently, as to give way or stop the car flow out of their pro-
grammed schedule. This is one challenge still to be solved for IoT and smart cities
development [3].

https://doi.org/10.1007/978-3-030-33229-7_22
Traffic Light Control Through Intelligent Agents, Sensors and Arduino 259
Cars in México are the main source of pollution which contributes to the formation
of tropospheric ozone, with serious consequences on cardiopulmonary illnesses, so the
government must not keep delaying the update of regulations for emissions at national
scale, as declared by officers of the Centro Mexicano de Derecho Ambiental AC
(CEMDA, Mexican Center for Environmental Rights), the CTS Embarq México, and
International Council of Clean Transportation (ICCT); these institutions remark that
México currently has very mild standards to control emissions of ozone precursors
pollutants [2].
Particularly, the city of Mérida, capital of the state of Yucatán, has become a city
where the automobiles are reducing spaces for the more than 2 million citizens and they
are flooding the streets of the historical downtown. According to the IMPlan, the
Municipal Institute for Mérida’s Planning, six hundred thousand private motor vehicles
are estimated in the city, that is to say there is one car for every two inhabitants in
average [4]. Since Mérida is a colonial capital city, with narrow streets, it is important
to find alternatives that help clear traffic congestions which are inevitably often pro-
duced in the main streets and surrounding areas downtown Mérida [5].
This paper describes a traffic control system based on intelligent agents, sensors and
Arduino to control traffic lights to improve every day traffic and emergency vehicles.
A scale model of a two streets intersection with traffic lights and sensors is built to
perform tests.
2 Reference Framework
This project involves intelligent agents, such as deliberative, reactive and information
agents; there are also four type of sensors which are key to show the conditions on the
streets and the system’s behavior during testing.
2.1 Intelligent Agents

An intelligent agent may be defined as a software entity which, based on its own
knowledge performs a set of operations aimed to satisfy the needs of a user or another
program, either by its own initiative or because of a user’s request [6].
All intelligent agents are programs but not all searching programs are intelligent
agents. An agent can be considered as an individual entity, sort of a part of a program
with control over its own functions; they are continuously performing processes that
tell them what to do and how to do it, they also communicate with other agents to solve
their given tasks adequately [7].
When a highly complex problem needs to be solve it is recommended to design a
multiagent system which is a system made of multiple intelligent agents able to fulfill
goals individually to solve a global goal [8].
According to [6], intelligent agents can be classified as deliberative, reactive,
information and hybrid agents. Deliberative agents are based on an internal symbolic
reasoning model which allows them to manage and coordinate tasks with other agents
according to action plans. This kind of agent uses a decision taking scheme based on
condition - cognition – action, to identify the environment conditions so that it can plan
the next step. There are also reactive or static agents whose functions are specific and
they are unable to perform any task out of those previously assigned; this does not
260 R. A. Ojeda-Cepeda et al.
imply that they are not able to interact with other agents to achieve specific joint
objectives. They do not save previous information therefore they are fast an efficient in
decision making but they do not adapt to change. Information agents are those that
have access to one or more sources of information and they are capable of filtrating or
manipulating data obtained from those sources in order to respond to user petitions or
to other information agents in cooperative environments. Information sources can be of
different nature from databases up to other agents. It is necessary to mention that a
particular implementation of an intelligent agent may show more than one behavior at
the same time, in which case it is classified as a hybrid agent. This allows to maximize
the strengths of some agents and minimize deficiencies of others, thus taking the best of
each type such as decision taking precision from the deliberative agents and speedy
response time from the reactive agents.
2.2 Related Work Review

Controlling traffic in the cities is an escalating problem with repercussions beyond the
trivial complication of waiting time and driver’s annoyance, and in a worldwide scope
there are pollution and smart cities development issues to consider. This is palpable
with the large selection of published related work from which we have only mentioned
those closer to the main subject of the present work. To improve waiting time
autonomous traffic control systems have been used with real time data reception to
control traffic lights and even to impose fines to reckless drivers [9], Q-learning
algorithm has been applied in [10] and in combination with particle swarm opti-
mization algorithm as in [11]. There are also proposals based on microcontroller
hardware devices which control infrared sensors and they adjust traffic light timing as
in [12] and [13], where the traffic light configuration is controlled via Bluetooth. Other
developments are focused on emergency vehicles to control traffic in an emergency
route by means hardware as in [14], and, in contrast there are proposals which privilege
pedestrians crossing of highways as in [15]. There are more elaborated systems that
incorporate fuzzy logic [16] and study cases involving air quality measurement [17]
with very interesting results in software simulations.
Some of the developments mentioned above share transit flow and traffic lights
sequences configuration with the present project; they take into account similar vari-
ables with variations in their definitions although mostly within the same range. Some
include other aspects such as fuzzy logic control and pedestrian crossing which are not
included in the present proposal. The scale model is the differentiator, which is a great
asset in a classroom environment and is quite useful to visualize testing results.
3 Design and Implementation
The system displays a multiagent architecture combining software and hardware. It is

built on a 0.80 cm by 0.80 cm wooden base with the programmed Arduido and sensors
mounted on a scale model of the street crossing with the traffic lights. Model size cars
can be moved about to see how the system reacts on each situation the user wishes to
test.
3.1 Architecture Design

The environment is perceived with a number of sensors connected to a programmed
Arduino with the three intelligent agent controlling the output, in this case three traffic
lights (actuators) on the streets. On the side of the software, the system consists of three
agents (see Fig. 1): a reactive agent which collects data from the sensors, a hybrid agent
(information and reactive agent) which gets, selects and sends information, and a
deliberative agent which takes decisions to control the traffic lights.
Fig. 1. Multiagent architecture (Color figure online)
The emergency agent is essentially a reactive one and it receives readings form the
color and sound sensors to determine the presence of emergency vehicles; it disregards
information related to traffic and time of the day, and it is sensible only to red or blue
lights. Once light is detected the emergency agent compares the sound levels on each
street to select the one with the highest noise so that such a street is identified as the one
with an emergency vehicle, (police patrol, ambulance or fire truck). Once an emergency
is detected and the street is tagged as that of preferential way, the emergency agent
sends data to the control agent which adds an information component to the former
agent and it is possible to classify the emergency agent as a hybrid one. The traffic
sensor agent receives readings from the distance and motion sensors which means it is
an information agent. All three distance sensors are identified individually and they
indicate if the line of cars waiting on each street has reached a certain point (where the
sensor is placed). In other words, these sensors alert when there is a traffic jam on each
street. Also, the motion sensors placed under each of the traffic lights count the passing
vehicles and this data is saved on a file for statistical purposes. Also, there is the
intelligent control agent which collects data from the emergency and traffic agents and
Fig. 2. Intelligent agent diagram
it decides the light sequence to be applied to the traffic lights in order to clear a traffic
jam or to give way to emergency vehicles. Figure 2 illustrates the described agents.
The scale model was built with the components described in detail as follows.
Arduino Mega 2560. A key component of the system is a programmable Arduino
Mega 2560 where the agents have been implemented and sensors are set to receive
information. Figure 3 shows the Arduino design with the following components.
Fig. 3. Arduino design (Color figure online)

Motion Sensor PIR HC-SR501. It detects when an object crosses directly in front of
the sensor and so it is programmed on the Arduino as a counting device. There are three
motion sensors to count vehicles passing on each street; this information shows how
many vehicles cross under the traffic lights after a light change. See number 1 on Fig. 3.
Ultrasonic Distance Sensor HC-SR04. This sensor continuously detects the distance
between its position and an object directly in front of it from 3 to 300 cm (see number 2
on Fig. 3). In this case, when consecutive readings remain constant and under 7 cm for
an allotted period of time it means that a car has remain in front of the sensor without
moving, hence it is interpreted as a jam on the street.
Sound Sensor KY038. It analogically detects the amount of generated noise in the
environment; the values range from 0 to 1023 (see number 3 on Fig. 3). The highest
value identifies the street where an emergency vehicle may be coming.
Color Sensor TCS230. This sensor is able to detect red, blue and green light around
it. To sense emergency vehicles it was programmed to react only to red and blue light,
because they are commonly used on such vehicles. The color sensor activates the sound
sensors on each street. See number 4 on Fig. 3.
3.2 Implementation
In Mérida, Yucatán, one of the busiest streets is the 50th st. which runs North to South;
at one point it is intersected by 95th street, which is a two-way road (see Fig. 4). This
intersection is represented on the scale model; it has 3 traffic lights, one is on 50th and
two more are on 95th. There are sound, distance, motion and light sensors installed on
this scenario. There are 3 ultrasonic sensors placed approximately at a distance of 3 to 5
average cars from the intersection. Traffic lights have a motion sensor installed
underneath each one. One light sensor on this scale model has been installed next to the
Fig. 4. Scale model of the streets intersection (Color figure online)

programmed Arduino as to have easy access to the sensor during testing. Sound sensors
are placed approximately at the same distance as the ultrasonic sensors on each street.
As it is still a prototype the whole system is sensible to malfunction of the sensors, in
which case the results will not be as expected.
4 Results
A number of tests were designed to verify the behavior of the agents on each set of
conditions (states) designed purposely to evaluate the results. Five typical states were
selected and they are explained in detail in the following paragraphs, according to the
corresponding 50th St, 95th East St and 95th West St. with a 10 s allocated time for each
traffic light on a regular basis.
1. Normal state. This state shows a normal sequence for the three traffic lights. It
starts with Traffic Light 1(TL1) allowing 8 cars to go for 10 s, followed by Traffic
Light 2 (TL2) allowing 5 cars for 10 s and finally Traffic Light 3 (TL3) allowing 3 cars
for 10 s, as detailed in Table 1. The count of cars is performed by each street´s motion
sensor. Since none of the distance, sound or color sensors are activated, the normal
sequence is followed.
Table 1. Normal state.

50th street 95th East street 95th West street
TL 1 D M S TL 2 D M S TL 3 D M S C
Stop Go (10 s) 5 cars Stop
Stop Stop Go (10 s) 3 cars
Go (10 s) 8 cars Stop Stop
2. Emergency state. When an emergency vehicle approaches on one road, the color
sensor (C) is activated by the red or blue light; the color sensors allows the sound
sensors to be evaluated in order to decide which street has an emergency vehicle
according to the highest sound sensor reading. Table 2 shows how the 95th East St is
cut short after allowing only 3 cars in 6 s on TL2 and then the sound sensors identify
50th street needing clearance for ambulance or patrol car; TL1 stays on for 20 s
allowing 6 cars to pass including the emergency vehicle. When that vehicle is far
Table 2. Emergency state.

Stop ✓ Go (6 s) 3 cars Stop ✓
Go (20 s) 6 cars Stop Stop
enough and the sound is no longer perceived, the traffic lights go back to a normal
schedule, as shown in TL3 with 10 s for 4 cars to pass on 95th West St.
3. Traffic jam state. In this context, a traffic jam is identified on a street with a long
line of unmoving cars according to the ultrasonic distance sensor value which will be
constant and less than 7 cm. after an allotted time. When such a situation arises, that
sensor requests a preference on the way to clear the blocked street, in other words, the
sensor requests a green light to be activated for as long as needed and when the
blockage is no longer sensed the system will go back to normal state. Table 3 illustrates
this situation, where TL2 allows 3 cars on a normal 10 s time slot; then it is TL3 turn
but two distance sensors on streets 50th and 95th East are activated, thus cutting down
to 8 s the time of TL3 to allow the system to decide which of the two jammed streets
should be served first: it gives way to the 50th street allowing 12 cars to circulate for
15 s while 95th East street needs to wait. After 50th street is cleared, 95th street is
granted a green light and a longer time slot of 12 s of green light was needed to allow 9
cars to go. Since both jammed streets are now cleared, the traffic light schedule goes
back to the normal state with 10 s on TL 3 with only 2 cars passing through.
Table 3. Traffic jam state.

Stop ✓ Stop ✓ Go (8 s) 0 cars
Go (15 s) 12 cars Stop ✓ Stop
4. Night state. From 00:01 h to 04:00 h, the night schedule is activated; this means
that the 50th street will have only flashing yellow light to yield or slow down and cross
the road carefully; on the other hand, the drivers on both ways of the 95th street will
have flashing red lights as to indicate drivers to be aware and cross after a full stop. The
night state disregards the distance sensor since no traffic congestions are expected
during late night. See Table 4.
Table 4. Night state.

Stop Yield Yield
5. Night state with emergency. During the night, TL1 stays on flashing yellow
mode and streets 95th East and West stay on red lights. As Table 5 shows, when an
emergency vehicle is identified on 95th East street, a green light is granted for as long
as needed (in this example it was 10 s) before going back to the preventive flashing
lights.
Table 5. Night state with emergency.

50th street 95th East street
95th West
street
Yield Stop Stop
Yield Stop ✓ Stop ✓
Stop Go (10 s) Stop
Yield Stop Stop
5 Final Remarks and Conclusions
The intelligent agent system devised for traffic light control on a scale model has been
successful to identify a set of conditions and it has reacted adequately according to each
situation. The control agent implemented on the Arduino allows sensors to perceive the
environmental variables to test the response of the multi agent system performance in
each situation, showing good decisions to regulate traffic flow on both streets. The
results show the system is able to change the environment favorably as traffic flow
concerns. The agent system implemented on the scale model is able to identify a traffic
jam and offers a way to free the blocked roads, as showed on state 3 Table 3, which
portrays an everyday situation in that particular street crossing in Mérida. Emergency
situations are recognized and handled adequately by the control agent thanks to the
sensors reading the environment in conjunction with the programmed Arduino.
The prototype was built using the Arduino which is easy to program using a variant
of C Language. This also allows the user to modify some parameters and test the results
on the scale model.
There is also the advantage of having extra ports in which case the system can be
scaled to model more sensors o traffic lights.
There is a lot to be done as future work on this incipient project: one line of study is
for this model to be a generator of patterns which can be the input for a more complex
traffic control learning system.
The authors would like to thank the Tecnológico Nacional de México, campus
Mérida, for the support and resources devoted to this work.
References
1. New York Times. https://www.nytimes.com/2019/01/21/upshot/stuck-and-stressed-the-
health-costs-of-traffic.html?ref=nyt-es&mcid=nyt-es&subid=article. Accessed 01 Aug 2019
2. Milenio. https://www.milenio.com/estados/autos-primera-fuente-de-contaminacion-en-el-
país. Accessed 01 Aug 2019
3. INEGI. http://www3.inegi.org.mx/contenidos/app/saladeprensa/aproposito/2015/trafico31.
pdf. Accessed 01 Aug 2019
4. Diario de Yucatán. http://yucatan.com.mx/merida/urbanismo/pobre-educacion-vial. Acces-
sed 16 Apr 2019
5. Diario de Yucatán. https://www.yucatan.com.mx/merida/uso-mas-eficiente-de-vehiculos.

Accessed 01 Aug 2019
6. Wooldridge, M., Jennings, N.: Intelligent agents: theory and practice. Knowl. Eng. Rev. 10
(2), 115–152 (1995). https://doi.org/10.1017/S0269888900008122
7. Inteligencia Artificial Agentes. https://sitiointeligenciaa.wordpress.com/agentes/. Accessed
01 Aug 2019
8. Universidad Nacional del Centro de la Provincia de Buenos Aires. http://www.exa.unicen.
edu.ar/catedras/optia/public_html/2018%20Agentes%20inteligentes.pdf. Accessed 01 Aug
2019
9. Universidad Nacional Abierta y a Distancia. https://stadium.unad.edu.co/preview/UNAD.
php?url=/bitstream/10596/17725/1/7702768.pdf. Accessed 01 Aug 2019
10. Rosyadi, A., Wirayuda, T., Al-Faraby, S.: Intelligent traffic light control using collaborative
Q-Learning algorithms. In: 2016 4th International Conference on Information and
Communication Technology (ICoICT), pp. 1–6. IEEE (2016)
11. El Hatri, Ch., Boumhidi, J.: Q-learning based intelligent multi-objective particle swarm
optimization of light control for traffic urban congestion management. In: 2016 4th IEEE
International Colloquium on Information Science and Technology (CiSt), pp. 794–799.
IEEE (2016)
12. Ghazal, B., et al.: Smart traffic light control system. In: 2016 Third International Conference
on Electrical, Electronics, Computer Engineering and Their Applications (EECEA),
pp. 140–145. IEEE (2016)
13. Universidad Autónoma del Estado de México. http://hdl.handle.net/20.500.11799/99060.
Accessed 01 Aug 2019
14. Kuzminvkh, I.: Development of traffic light control algorithm in smart municipal network.
In: 2016 13th International Conference on Modern Problems of Radio Engineering,
Telecommunications and Computer Science (TCSET), pp. 896–898. IEEE (2016)
15. Universidad Politécnica de Madrid. http://oa.upm.es/48212/1/TFG_SILVIA_MONINO_
ESTEBAN.pdf. Accessed 01 Aug 2019
16. Orozco, H., Lazcano, S., Landassuri, V.: Simulación basada en agentes para el control
inteligente de semáforos mediante lógica difusa. In: Pistas Educativas No. 128 (SENIE
2017), México, pp. 1206–1223 (2018)
17. Scientific Electronic Library. http://www.scielo.org.mx/pdf/poli/n50/n50a10.pdf. Accessed
01 Aug 2019
A Clustering-Based Approach
to Base Station Assignment
in IoT Cellular Systems
Edgar Adrian Esquivel-Mendiola1(B) , Hiram Galeana-Zapién1 ,

and Edwin Aldana-Bobadilla2
1
Centro de Investigación y de Estudios Avanzados del I.P.N., Unidad Tamaulipas,
Ciudad Victoria, Mexico
{eesquivel,hgaleana}@tamps.cinvestav.mx
2
Conacyt-Centro de Investigación y de Estudios Avanzados del I.P.N.,
Unidad Tamaulipas, Ciudad Victoria, Mexico
ealdana@tamps.cinvestav.mx
http://www.tamps.cinvestav.mx
Abstract. We address the optimal base station assignment for each

device in IoT networks. In this regard, we extend the well known k-
means clustering algorithm wherein each device is represented as a n-
tuple which encompasses the channel conditions for a set of candidates
base stations to be assigned. Our solution firstly computes the number of
clusters in the scenario, and then determines the objects (devices) belong-
ing to each cluster (group). Simulations results show that our approach
achieves competitive results in terms of the average sum throughput and
load balancing between the cluster heads.
Keywords: Base station assignment · Clustering · Internet of Things
1 Introduction
Cellular communication systems will play a key role in the provision of services
in the Internet of Things (IoT) systems. This is particularly critical in scenarios
where mobile sensing objects are expected to be deployed over large areas (e.g.,
logistics, fleet management, etc.), requiring the support of cellular networks to
seamlessly transmit/receive information to/from Internet. However, the support
of IoT imposes several challenges relative to the cellular communication systems,
such as: (a) the expected number of connected devices per access point or base
station (BS), leading to a significant traffic increase per service area [1]; (b) het-
erogeneity of devices and quality of service (QoS) requirements; and (c) resource
constraints either in the device (e.g., energy, computation) or network side (e.g.,
resource availability) [2]. In order to efficiently support IoT in cellular networks,
resource management strategies are required which in turn must guarantee QoS
requirements of connected devices, as well as limiting interference and tackling
new resource bottlenecks like the backhaul [3–5].
https://doi.org/10.1007/978-3-030-33229-7_23
Clustering for BSA Problem 269
In this context, one of the most challenging problem is to find that assignment
of BS that optimizes radio transmission to/from mobile devices. This poses an
optimization problem where the objective function is defined typically in terms
of utility and cost functions associated to a potential assignment. Such an opti-
mization problem can be subject to a set of constraints (i.e., device and system
capacity constraints) which in turn can induces non-convex solution spaces, in
which case several authors resort to heuristic approaches [CITE]. For instance,
given the heterogeneous characteristics of IoT devices, some scenarios poses con-
straints in terms of computation, sensing, and communication capabilities.
Implicitly the assignment of BS induces a partition with subsets that include
devices with “similar” features wherein the similarity can be defined in terms
of a metric (e.g. relative to channel conditions). This assumption allows us to
treat such a problem as a clustering problem. This approach has been explored
in [6], where the similarity metric is based exclusively on a proximity measure
between devices (Euclidean distance). This implies that the approach in groups
nearby devices without taking into account channel conditions (i.e., slow and fast
fading), and thus the resulting BS assignment might not be the most appropriate
from the air interface performance point of view.
In order to define a clustering-based BS assignment, we define a device rep-
resentation which is a scalar value that denotes the overall channel conditions
of each device with respect to the potential candidate set of BSs. Such represen-
tation is needed to be used as criterion to group devices according to channel
conditions instead of Euclidean distance. Furthermore, we focus on IoT cellular
systems, in which devices are expected to be connected to the Internet either via
a macro-cell BS or by means of a cluster heads acting as the serving BS or Sink.
In this work we concentrate in this latter case to determine the BS assignment
solution (i.e., the associations between IoT devices and cluster heads), where
a given mobile and the cluster head is supported via device-to-device (D2D)
communications.
1.1 Contributions
This work provides the following significant contributions:
– Algorithms from data mining cannot be applied directly to group devices in

cellular systems, given that the attributes of devices are mainly related to
the spatial position in the cell. Thus, we provide a device representation that
accounts for the channel conditions (i.e., path-loss, shadowing) of devices to
group them using the well-known k-means algorithm.
– The proposed clustering-based BS assignment approach allows to control the
load among cluster heads, so that when a cluster head exceeds a given load
threshold our approach triggers a process to select a device to become a
cluster head to have a new group of users.
270 E. A. Esquivel-Mendiola et al.
– In conventional k-means implementations, an input requirement to the algo-

rithm is the number of cluster heads (i.e., k, the number of groups). In this
sense, our approach allow us to compute the appropriate number of clus-
ter heads that improve the system capacity and provide balance conditions
among cluster heads.
– The proposed approach can operate in a distributed way as each cluster head
decide separately which devices are accepted according to load conditions.
The remainder of this paper is organized as follows. In Sect. 2 the related

work is presented. The Sect. 3 introduces a reference BS assignment framework,
system model and the formalization of the BS assignment problem. Then, in
Sect. 4, we detail the proposed clustering-based BS assignment algorithm for
which a suitable devices’ representation of devices is described. The simulation
experiments along with numerical results are discussed in Sect. 5. Finally, the
conclusions remarks are summarized in Sect. 6.
2 Related Work
The BS assignment problem has been extensively studied in the literature, where
greedy to heuristic-based algorithms have been proposed so far. On the one hand,
in greedy solutions, the devices are associated with the BS that offers the highest
signal to noise ratio (SNR). On the other hand, for heuristic approaches the
problem is formulated as an optimization problem, imposing different resource
constraints [7–10]. In [7], the authors formulated the BS assignment problem
as the Multiple-Choice Multidimensional Knapsack Problem and proposed a
heuristic approach to solve the problem. In [11] the authors studied the joint
BS assignment and resource allocation problem, and proposed solution based on
dynamical programming. Similarly, in [9] the authors modeled the problem as
a supply and demand scheme. Game theoretic-based solutions have also been
applied in [10] leading to a distributed-based assignment decision.
Recently, there is a trend to study the design of clustering-based solutions for
the BS assignment problem in cellular networks [12–14]. This is mainly boosted
by the supporting of device-to-device communications, where devices are mod-
eled to communicate directly with the minimum coordination of the BS [15–17].
Most of the approaches proposed so far are based on the extensively work pre-
sented for wireless sensor networks, where clustering solutions have been pro-
posed to perform the grouping of devices and the selection of the sink nodes
[18,19]. In particular, the attributes of devices are determined by their spatial
position inside the cell, and a clustering algorithm determines the membership
of a device to one group. The criterion to decide the clusters of devices is mainly
based on the Euclidean distance. In [20], the authors studied the support of
machine-to-machine communications under cellular networks scenarios and pro-
posed a method for grouping the devices and cluster head selection. In [13]
a clustering approach is studied for the joint BS assignment and load balanc-
ing problem. Authors in [6] proposed a cluster head selection and a clustering
scheme based on k-means. Similarly, in [21], the authors proposed a cluster head
selection method based on resource availability at devices. In [22], the authors
proposed a clustering solution based on physical position and social interaction
metrics between devices. The work in [23] proposed a method to group devices
that are configured to operate in D2D mode. Similarly, an energy-constrained
scenario is studied in [24], where the selection of cluster heads is determined by
the energy budget of devices.
It is worth noting that in most of the analyzed works the attributes of devices
are characterized by their position in the cell. However, this representation does
not account for the channel conditions observed between devices and their poten-
tial cluster head candidates. Moreover, devices are associated with nearest cluster
head according to the similarity function based on the Euclidean distance, which
may lead to a poor system performance since devices may not be associated with
an adequate cluster head that satisfies their minimum requirements.
3 Reference Framework
In order to evaluate the BS assignment problem, herein we define a generic single-

cell cellular system. Yet simple, the presented reference framework allow us to
design and to validate a clustering-based algorithm to solve the BS assignment
problem. Specifically, we assume that there are M total devices camped in the
service area of a single-cell coverage. Each device i ∈ M could be considered as
a cluster head mode with C ⊂ M or an IoT device with D ⊂ M , but not both
modes at the same time. As depicted in Fig. 1, we assume that there exist the
following type of devices:
– IoT devices. These devices are assumed to be resource-limited nodes that

require to be connected to a cluster head for downlink and uplink data trans-
fer. Each IoT device can only be connected to one cluster head (i.e., no
partial associations are considered), which is defined as a device-to-cluster
(D2C) association. Moreover, each IoT device requires to meet a minimum
data rate requirement in the downlink, denoted as Rimin .
– Cluster heads. Devices acting as cluster heads are directly connected to
the macro-cell BS (C2N). As cluster heads provide connectivity to a given
number of IoT devices, they are assumed to be constrained with an amount
of radio resources.
At this regard, in this paper we concentrate on the problem of allocating

IoT devices to a given cluster head. For the sake of simplicity, we refer to this
problem as BS assignment problem.
Fig. 1. Generic hierarchical single-cell scenario. Cluster heads are assumed to be con-
nected to the macro-BS, whereas the association of IoT devices to a cluster head is
determined using a clustering-based approach.
3.1 System Model

All devices can be selected as cluster heads. The suitability of a given device i to
become a cluster head is determined in terms of the signal to noise ratio (SNR).
Thus, we compute the SNR between each device with respect to each of the rest
of devices. Hence, the computation of the SNR achieved by device i being served
by cluster head j (with i = j) is computed as:
Gij Pij
SN Rij = (1)
η
where Gij accounts for the radio channel gain between device i and device j
(including path-loss, shadowing and noise figures), Pij is the transmit power of
device j being in cluster head mode to device i, η denotes the thermal noise.
Taking this into account, the achievable transmission rate of device i assigned
to cluster head j (i.e.. downlink direction), denoted by rij , is determined by:
rij = W log2 (1 + SN Rij ) (2)

where W is the system bandwidth assuming a frequency reuse factor of 1. Fur-
thermore, in order to quantify the resource consumption of device i assigned to
the cluster head j, we define a radio resource cost function, denoted as αij .
Rimin
αij = (3)
uij
where uij is the utility achieved by device i assigned to device j as a function of
the transmission rate rij achieved by device i, and αij = 1 would mean that the
assignment of device i to device j requires all available radio resources to meet

its rate requirement.
3.2 BS Assignment Problem
As illustrated in Fig. 1, the BS assignment problem aims to assign each IoT device
to a given cluster head in order to optimize the system’s throughput. We assume
that IoT devices have a minimum downlink data rate requirement, whereas the
cluster heads are assumed to have a limited amount of radio resources. Over
such a basis, we define B as the assignment matrix of D × C elements, where
each element bij is determined as follows:

1, if device i is assigned to cluster head j
bij = (4)
0, otherwise
Therefore, the optimization problem is defined as follows:
C
D
max uij bij (5a)
bij
j=1 i=1
M

s.t. αij bij ≤ 1, j = 1, . . . , C (5b)
i=1
C

bij = 1, i = 1, . . . , D (5c)
j=1
Ri ≥ Rimin , i = 1, 2, . . . , D. (5d)
bij ∈ {0, 1} (5e)
The BS assignment problem (5a) is a combinatorial problem due to the inte-

ger variables. Constraint (5b) works as a cost capacity for each cluster head, that
also could be applied to denote the load condition of cluster head. Constraint
5c states that all IoT devices must be assigned to a single cluster head. Con-
straint 5d ensures that the minimum rate requirement of all devices is satisfied.
Finally, constraint 5e indicates that the assignment variable only can take binary
variables.
4 Clustering-Based BS Assignment Algorithm
In this section, we describe our clustering-based BS assignment algorithm. In

what follows we firstly introduce the proposed device characterization.
(a) Candidates for device i
(b) Devices represented by channel features
Fig. 2. Characterization process of IoT devices in cellular networks: (a) Candidate set
of BSs for each user, and (b) each user characterized in terms of SNR and path loss
attributes.
4.1 Clustering Characterization

In order to apply clustering algorithms over a set of objects, we need to firstly
represent the objects by means a set of two or more attributes. Then, a measure
of similarity is applied to group the objects that share the same attributes.
Unlike related works, which resort on spatial attributes when addressing the
BS assignment problem as a clustering problem, we define a representation that
accounts for the SNR and path loss conditions observed between a given device
and its potential BS candidates.
In order to cope with the problem stated before, we firstly compute for each
device i the SNR, path-loss and cost observed to/from its candidates. The result-
−−−−→ −−→ →
ing metrics are represented in a vector denoted by SN Ri , P Li , −
αi . As showed in
Fig. 2(a), the set of candidates for device i is determined by all devices that are
between a path loss margin. Furthermore, in order to determine the attributes

of each device according to its observed channel conditions, we compute a scalar
value that represents the average metric between each device and its candidates,
as follows:
|CS|
1
SN Ri = SN Rij (6)
|CS| j=1
where SN Ri denotes the average SN R observed between device i to/from its

candidates. The same procedure is repeated to calculate the average path loss
and cost between device i and each device on its candidate set. CS denotes
the set of candidates of device i for each metric. As a result, devices are now
represented by their attributes computed from channel conditions, as depicted in
Fig. 2(b). The aim of the proposed representation is to account for the channel
conditions observed by each device, which can be used by a given clustering
algorithm to find the groups of devices belonging to each cluster head. In the
following we detail our proposed clustering algorithm, which is an extension of
the k-means algorithm.
The proposed clustering-based algorithm consists of two phases, namely clus-
ter heads searching and modified k-means algorithm. The former follows a load
based criterion to determine the number cluster heads, while the latter one is an
extension of the well-known k-means algorithm to perform the clustering pro-
cess. Firstly, the cost metric between each device and all devices in the system is
computed. This is necessary since all devices could be selected as cluster heads.
Then, k devices are randomly selected as the initial cluster heads (step 4), and
the rest of devices are grouped with the cluster head j that satisfy the maximum
M
SNR criterion (steps 5–8). After that, the total radio cost πj = i=1 αij bij of
each cluster head j is computed. The total cost of each cluster head πj could
be seen as a measure of the load condition observed as a result of the grouping
process. At step 13, if the cost capacity of each cluster head is not exceeded, the
number of k is adequate to perform the clustering process. Otherwise, the algo-
rithm continue with search for the number of cluster heads. Each cluster head
exceeding cost capacity mark its associated devices with high cost as rejected
devices, which is denoted by RD. That is, each cluster head accepts the devices
to which its sum cost satisfy the condition πj < 1. Also, the assignment variable
is updated with bij = 0, and each device marks j as a visited cluster (step 17).
Next, each device tries to be assigned to another cluster head that has not been
visited, and the assignment variable is updated. This process is repeated (steps
14–26) until each device has tried to be associated with all the cluster heads,
or the assignment does not exceed the cost capacity in any cluster head. At the
end, it could be possible that some devices may not be assigned to any cluster
head, so there is checked if the RD set is not empty (step 28). If there are devices
not assigned, one new cluster head is determined by choosing randomly a device
from the RD set. The device is added to the set of cluster heads and step 5 is
repeated.
Once the number of k has been found in the first phase, the process continue
to execute the k-means algorithm. Firstly, k devices are selected randomly to
be the initial centroids. Then, devices are associated with the closest cluster
head according to the maximum SNR (step 38). The centroids are updated by
calculating the mean of the devices in each cluster (step 39). This process is
repeated until no more changes are observed in the cluster centroids.
5 Simulation Experiments
In this section, we study the performance of the proposed clustering-based base
station assignment algorithm. Firstly, we developed a custom made simulator
in MATLAB to simulate the network scenario, which is composed of a single
macro-BS at the center of the cell with a radius of 500 m. Then, IoT devices
are uniformly distributed over the coverage area of the cell keeping a separation
distance of 125 m between devices and the macro-BS. The maximum transmit
power of the macro-BS is set to 47 dBm, whereas cluster heads have a transmit
power of 18 dBm. A path-loss margin of 6 dbm is selected to define the can-
didate set of each device. Moreover, all devices are assumed to have the same
downlink data rate requirement Rimin . Log normal shadowing is accounted with
standard deviation of 8 dB. The rest of simulation parameters are summarized
in Table 1. We compare the proposed clustering-based approach with the tra-
ditional k-means algorithm. For the k-means, the spatial position of the IoT
devices is considered as the attribute during the clustering process. Moreover,
an arbitrary value of k selected as the number of clusters for the k-means algo-
rithm. On the contrary, in our approach, we firstly determine the number of k
in the first stage of the algorithm, and then the k-means algorithm performs the
grouping process.
5.1 Simulation Results
Results are divided into clustering analysis and performance evaluation of the
proposed clustering-based solution. The former consists in the analysis of the
average number of cluster heads and devices per cluster head. The latter consist
in a comparative between the proposed algorithm and the conventional k-means
algorithm.
Tables 2 and 3 show the average number of clusters and the average number
of device per clusters. The comparative was made for a variation of devices in the
system and minimum data rate requirement ranging from 15 to 100 kbps. From
the results there should be noted that as the rate requirement and the number of
devices increases, the number of formed cluster heads also increases. In addition,
there is observed that the average number of devices per cluster head remains
equal for the variation in the number of devices and the rate requirement.
Algorithm 1. Cost-based clustering algorithm

Data: P Lij ,SN Rij
Result: Set of CHs C,assignment matrix B
1 Cluster heads searching
2 forall the i ∈ M , j ∈ M do
3 Compute αij , and uij
4 end
5 Choose k devices randomly as CHs
6 for each device i find the most valuable CH j ∗ do
7 j ∗ =arg max(SN Rij )
8 bij ∗ = 1
9 end
10 for each CH j compute total radio resource cost do

11 πj = M i=1 αij bij
12 end
13 if πc > 1 for any h then
14 while CH not visited do
15 for each CH j with πj > 1 do
16 Find devices exceeding cost capacity and mark them as rejected
devices RD
17 For each device i ∈ RD set bij = 0 and mark CH j as visited
18 end
19 for each device i in RD do
20 Select most valuable (not visited) CH j ∗
21 bij ∗ = 1
22 end
23 for each CH j compute total radio resource cost do

24 πj = M i=1 αij bij
25 end
26 end
27 end
28 if RD not empty then

29 Choose randomly a device i from RD to create a new CH

30 i ∪C
31 Repeat step 5 until all devices are assigned
32 end
33 end
34 Modified k-means algorithm
35 Given the number of CHs previously found, set k = |C|
36 Choose randomly k devices from M as the initial centroids
37 while no change in centroids do
38 Assign device i to the closest CH j based on the SN Rij
39 Recalculate the new centroid for each cluster
40 end
41 end
Table 1. Simulation parameters.
Parameter Value
Cell radius 500 m
Operating frequency 2000 MHz
Bandwidth Macro: 10 MHz
Macro Path-Loss 128.1 + 37.6log10 (distance km)
D2C Path-Loss 148 + 40log10 (distance km)
max
Macro-BS transmit power PBS 46 dBm
Cluster head transmit power PCmax 18 dBm
Table 2. Average number of cluster heads.
Devices Rate requirement Rmin

15 Kbps 20 Kbps 25 Kbps 30 Kbps 40 Kbps 60 Kbps 80 Kbps 100 Kbps
50 6.692 8.43 10.13 11.616 14.498 19.234 25.556 26.96
100 12.382 15.772 18.996 22.024 27.52 36.486 49.854 52.598
150 17.86 22.882 27.6 32.076 40.318 53.562 73.844 77.898
200 23.206 29.894 36.146 42.064 52.964 70.344 97.044 103.25
250 28.564 36.556 44.448 51.822 65.282 87.346 119.598 128.366
Table 3. Average number devices per cluster head.
Devices Rate requirement Rmin

15 Kbps 20 Kbps 25 Kbps 30 Kbps 40 Kbps 60 Kbps 80 Kbps 100 Kbps
50 6.522 4.955 3.955 3.320 2.461 1.607 0.958 0.857
100 7.101 5.357 4.275 3.549 2.640 1.744 1.006 0.902
150 7.418 5.569 4.444 3.682 2.724 1.802 1.032 0.926
200 7.635 5.702 4.542 3.760 2.778 1.844 1.062 0.937
250 7.769 5.849 4.632 3.829 2.831 1.863 1.093 0.947
Devices Distribution Between Cluster Heads: Figure 3 shows a graphical

comparative in terms of the assignment found by our proposed clustering-based
solution and the k-means algorithm. Initially, a value of k = 5 was set for both
algorithms; however, as showed in Fig. 3(b) the number of cluster heads found
by our proposed approach was of k = 24. This is because our approach follows
a load balance criterion which creates a new cluster head when imbalance is
observed. Moreover, there should be noted that the k-means algorithm groups
the devices according to their spatial position then devices are assigned to the
closest cluster head, whereas our proposed approach groups the devices according
to the representative metrics of SNR and cost.
Fig. 3. Comparative of the assignment solution found by the k-means and our pro-
posed clustering-based approach. The number of devices was set to 150 with a rate
requirement of 25 Kbps.
Analysis of System Throughput: Figure 4 shows the cumulative distribution

function of the achieved SNR by each algorithm for different minimum data rate
requirements of devices. As shown in Fig. 4, our proposed clustering-based solu-
tion has a good performance as the percentage of devices achieving a good SNR
value is greater than the achieved with the k-means. For instance, in Fig. 4(a)
the 50% of the devices achieved an SNR value of 95 dBm with the k-means
algorithm, while the achieved SNR with our clustering-based approach was of
105 dBm. A similar behavior is observed as the data rate requirement of devices
is increased.
Fig. 4. Cumulative distribution function (CDF) for different data rate requirements.
The number of devices is set to M = 150, and k = 5 for the k-means algorithm.
Analysis of the Average Sum Cost: Figure 5 shows the average sum cost per
cluster head achieved by each algorithm under the variation of the number of
devices. There should be noted that in our proposed approach, we have defined
a cost constraint, which allows us to balance the load conditions between the
cluster heads. From Fig. 5, the average sum cost of our proposed approach is
below one, whereas the assignment solution found by the k-means algorithm
always exceeds the maximum capacity. For instance, in the case of 250 devices,
the k-means algorithm has an average cost of six times the maximum allowed
capacity at each cluster head. The reason behind is that the k-means algorithm
has a fixed value k which leads to overhead in clusters causing a cost increase.
Fig. 5. Average sum cost for each cluster head as the number of devices is varied. The
devices’ data rate requirement is also varied ranging from 15 to 30 kbps.
6 Concluding Remarks
In this paper, a novel clustering-based base station assignment algorithm has
been proposed to support the deployment of IoT devices under cellular com-
munication systems. Unlike other clustering approaches, in our solution, we
have developed a device representation that accounts for the channel conditions
between the device and its potential candidates. We have shown through exper-
iments that our proposed approach has significant gains in terms of the average
sum throughput. Moreover, in terms of the average sum cost between the cluster
heads, our approach has a better performance than the k-means algorithm. As
future work, we aim to improve the process of the cluster heads discovery by
exploring combinatorial approaches. In addition, we aim to study the behavior of
other clustering algorithms from data mining for examining the behavior under
the massive presence of IoT devices.
References
1. Cisco Systems: Cisco visual networking index: Global mobile data traffic forecast
update, 2017–2022 white paper. Technical report, Cisco Systems, February 2019
2. Dhillon, H.S., Huang, H., Viswanathan, H.: Wide-area wireless communication
challenges for the internet of things. IEEE Commun. Mag. 55(2), 168–174 (2017)
3. Ge, X., Tu, S., Mao, G., Wang, C., Han, T.: 5G ultra-dense cellular networks.
IEEE Wirel. Commun. 23(1), 72–79 (2016)
4. Jafari, A.H., López-Pérez, D., Song, H., Claussen, H., Ho, L., Zhang, J.: Small
cell backhaul: challenges and prospective solutions. EURASIP J. Wirel. Commun.
Netw. 2015(1), 206 (2015)
5. Cheng, L., Gao, Y., Li, Y., Yang, D., Liu, X.: A cooperative resource allocation
scheme based on self-organized network in ultra-dense small cell deployment. In:
2015 IEEE 81st Vehicular Technology Conference (VTC Spring), pp. 1–6, May
2015
6. Zhao, Y., Liu, K., Xu, X., Yang, H., Huang, L.: Distributed dynamic cluster-head
selection and clustering for massive IoT access in 5G networks. Appl. Sci. 9(1),
132 (2019)
7. Galeana-Zapien, H., Ferrus, R.: Design and evaluation of a backhaul-aware base
station assignment algorithm for OFDMA-based cellular networks. IEEE Trans.
Wirel. Commun. 9(10), 3226–3237 (2010)
8. Dhillon, H.S., Andrews, J.G.: Downlink rate distribution in heterogeneous cellular
networks under generalized cell selection. IEEE Wirel. Commun. Lett. 3(1), 42–45
(2014)
9. Shen, K., Yu, W.: Downlink cell association optimization for heterogeneous net-
works via dual coordinate descent. In: 2013 IEEE International Conference on
Acoustics, Speech and Signal Processing, pp. 4779–4783, May 2013
10. Semiari, O., Saad, W., Bennis, M.: Downlink cell association and load balanc-
ing for joint millimeter wave-microwave cellular networks. In: 2016 IEEE Global
Communications Conference (GLOBECOM), pp. 1–6, December 2016
11. Lai, W.K., Liu, J.: Cell selection and resource allocation in LTE-advanced hetero-
geneous networks. IEEE Access 6, 72978–72991 (2018)
12. Hajjar, M., Aldabbagh, G., Dimitriou, N., Win, M.Z.: Hybrid clustering scheme for
relaying in multi-cell LTE high user density networks. IEEE Access 5, 4431–4438
(2017)
13. Kollias, G., Adelantado, F., Verikoukis, C.: Spectral efficient and energy aware clus-
tering in cellular networks. IEEE Trans. Veh. Technol. 66(10), 9263–9274 (2017)
14. Kazmi, S.M.A., Tran, N.H., Ho, T.M., Manzoor, A., Niyato, D., Hong, C.S.: Coor-
dinated device-to-device communication with non-orthogonal multiple access in
future wireless cellular networks. IEEE Access 6, 39860–39875 (2018)
15. Tehrani, M.N., Uysal, M., Yanikomeroglu, H.: Device-to-device communication in
5G cellular networks: challenges, solutions, and future directions. IEEE Commun.
Mag. 52(5), 86–92 (2014)
16. Vlachos, C., Friderikos, V.: Optimal device-to-device cell association and load bal-
ancing. In: 2015 IEEE International Conference on Communications (ICC), pp.
5441–5447, June 2015
17. Xiao, S., Zhou, X., Feng, D., Yuan-Wu, Y., Li, G.Y., Guo, W.: Energy-efficient
mobile association in heterogeneous networks with device-to-device communica-
tions. IEEE Trans. Wirel. Commun. 15(8), 5260–5271 (2016)
18. Rostami, A.S., Badkoobe, M., Mohanna, F., Keshavarz, H., Hosseinabadi, A.A.R.,
Sangaiah, A.K.: Survey on clustering in heterogeneous and homogeneous wireless
sensor networks. J. Supercomputing 74(1), 277–323 (2018)
19. Sarkar, A., Murugan, T.S.: Cluster head selection for energy efficient and delay-less
routing in wireless sensor network. Wirel. Netw. 25(1), 303–320 (2019)
20. Tu, C., Ho, C., Huang, C.: Energy-efficient algorithms and evaluations for massive
access management in cellular based machine to machine communications. In: 2011
IEEE Vehicular Technology Conference (VTC Fall), pp. 1–5, September 2011
21. Fodor, G., Parkvall, S., Sorrentino, S., Wallentin, P., Lu, Q., Brahmi, N.: Device-
to-device communications for national security and public safety. IEEE Access 2,
1510–1520 (2014)
22. Wang, L., Araniti, G., Cao, C., Wang, W., Liu, Y.: Device-to-device users clustering
based on physical and social characteristics. Int. J. Distrib. Sens. Netw. 2015(8),
1:1 (2015)
23. Koskela, T., Hakola, S., Chen, T., Lehtomaki, J.: Clustering concept using device-
to-device communication in cellular system. In: 2010 IEEE Wireless Communica-
tion and Networking Conference, pp. 1–6, April 2010
24. El-Feshawy, S.A., Saad, W., Shokair, M., Dessouky, M.I.: An efficient cluster-
ing design for cellular based machine-to-machine communications. In: 2018 35th
National Radio Science Conference (NRSC), pp. 177–186, March 2018
Fuzzy Logic Controller for Automation
of an Autonomous Irrigation System Designed
for Habanero Pepper
(Capsicum Chinense Jacq.)
Jeni Molina-Puc, Mauricio Gabriel Orozco-del-Castillo(&),

Dakar Fernando Villafaña-Gamboa, Romeo Alam Gómez-Buenfil,
José Misael Guzmán-Tolosa, and Humberto Sarabia-Osorio
Departamento de Sistemas y Computación,

Tecnológico Nacional de México/I.T. Mérida, Mérida, Mexico
{e15081607,mauricio.orozco,dakar.villafana,e15080988,
e15080844,e16080414}@itmerida.edu.mx
Abstract. Agriculture is a fundamental economic activity but implies a large

consumption of water; in Mexico, the habanero pepper (Capsicum chinense
Jacq.) is one of the main agricultural products. This work presents the design of
a Sugeno based fuzzy controller with 18 if-then rules for an irrigation system
that allows irrigation to be managed in a sustainable manner, reducing the loss
of water giving habanero pepper plants a favorable environment for their
growth. The controller consists of a set of rules designed to activate or deactivate
the irrigation system based on four input variables.
Keywords: Fuzzy logic Intelligent irrigation system Capsicum chinense

Jacq Control systems
1 Introduction
The origin of irrigation systems goes back thousands of years when the rivers were
used to irrigate the crops using a staircase design [1]. However, nowadays there is a
water crisis due to the misuse that is made of it, causing the environment and millions
of people to suffer the consequences [2]; according to the World Water Council, there
are six major problems related to water: (1) scarcity, (2) lack of access, (3) decrease in
quality, (4) decrease in the allocation of financial resources, (5) fragmentation in the
administration and (6) the lack of awareness of the people when making an inadequate
use of the water [3]. The Organization of the United Nations and Agriculture reports
that 70% of the world’s water is used mainly for activities such as agriculture [4].
Habanero pepper (Capsicum chinense Jacq.), after the tomato, is one of the main
products of agriculture, grown in the states of Baja California Sur, San Luis Potosi,
Sonora, Tabasco and in the states that form the Yucatan peninsula: Campeche,
Quintana Roo and Yucatan, which have the denomination of origin of habanero pepper
since 2010 [5]. The traditional cultivation techniques for this plant are obsolete, this is
because the producers are limited in technology, income, among others [6]. According
https://doi.org/10.1007/978-3-030-33229-7_24
Fuzzy Logic Controller for Automation of an Autonomous Irrigation System 285
to [7], the cultivation of habanero peppers is generally in the open field and with
manual irrigation, which ends up affecting production by diverse environmental fac-
tors, capable of reducing the quality of the product, yield and profitability of the
crop. Nowadays, the way to improve the crop has been sought using a controlled area
where external factors are reduced [8].
Traditional farming techniques such as conventional irrigation systems have
become obsolete due to limited technology and lack of income from producers. This is
why irrigation systems have been updated to date by using automatic, autonomous and
even intelligent controllers in order to improve the crop and reduce the physical work
of farmers to a level where they only are required to perform a supervision of the
system instead of a manual control of it [9].
Fuzzy logic is a branch of artificial intelligence that refers to the way people
describe the environment that surrounds them, so it has turned out to be successful
when applied to problems related to uncertainty and vagueness [10]. There are con-
trollers based both on classic crisp sets and fuzzy sets. Crisp sets are sets with a discrete
border and are an important tool for mathematics and science, however, when used in
control applications, they do not demonstrate the thinking of the human being or the
way in which he or she interprets the surrounding environment. Crisp sets are defined
as a collection of objects or elements, which can belong to the set or not (true or false,
respectively); there are no intermediate values [11]. On the contrary, a fuzzy set does
not have a discrete border, so that belonging or not to some set occurs gradually [12].
The fuzzy set is an extension of the classical set, to which a defined membership
function is added with a range between zero and one. A fuzzy set allows then to
simulate the process of human reasoning to control systems based on their experience
and solve problems using mathematical methods based on degrees of membership or
belonging, associating them with linguistic values, defined with words or labels used
naturally by a human being. When modeling information using statements that are not
totally true or totally false, they are applied to expressions that can take a truth value
from a set of values composed of two extremes and all their possible intermediate
values. This modeling is carried out by using linguistic rules that approximate functions
of inputs and outputs of the system [11].
Incorporating fuzzy logic with control systems results in fuzzy control systems or
fuzzy inference mechanisms, which are responsible for determining what should be
done to fulfill a process in the best possible way based on knowledge and experiences
provided by a human being in charge of operating the system, since without the
knowledge base it is impossible to develop a controller that works correctly [13].
A system based on fuzzy logic necessarily requires the use of different concepts such as
fuzzy set, membership function, rules, linguistic variables and values [14]. The
membership function is responsible for representing the degree of belonging of an
element to a set, that is, represents the degree to which an element is part of such a set
[14]. There are different ways to represent membership functions, among the most
common are Gaussian, triangular, singleton, trapezoidal and sigmoidal. If-then rules are
used and form a fundamental component in an inference system that can represent
human experience in a specific process of a controller [14].
A fuzzy inference system (FIS) is a computational model based on fuzzy sets, rules
and reasoning. To develop a FIS three conceptual components are required: (1) a set of
286 J. Molina-Puc et al.
rules with a selection of fuzzy rules, (2) a database in which the membership functions
to be used in the fuzzy rules are defined, and (3) a reasoning mechanism. There are
basic inference systems in which you can have fuzzy and discrete inputs, but the output
must always be fuzzy, however, sometimes it is necessary to have a discrete output, as
in the case of controllers. In these cases, it is necessary to have a defuzzification method
to obtain a discrete value that best represents the fuzzy set. The different types of fuzzy
inference systems are Mamdani, Sugeno and Tsukamoto; the type of system that should
be used depends primarily on the nature of the output [11].
Currently, there are works that implement fuzzy logic in control modules, which
logically determine the solutions to achieve the control objectives from the knowledge
that is provided by the human being [11]. In [15] the operation of a fuzzy controller
applied to an irrigation system is shown; the controller uses triangular and trapezoidal
membership functions for the fuzzification of the variables and a rule base of the
Mamdani type. In [16] a fuzzy logic controller is used, also based on the Mamdani
model for the automation of a greenhouse irrigation system, taking into account as the
input value the difference between the desired soil moisture and the actual soil mois-
ture, and as output value a valve that allows to open and close the water passage; this is
done to effectively estimate the amount of water avoiding that too much water is
applied to the plants. In [17] an irrigation system based on the information provided by
humidity, temperature and soil moisture sensors is proposed; considering the inputs, the
output of the system determines the irrigation time for the crop while optimizing water
use. In [18] and [19] climatic factors are used to determine the ignition of the irrigation
pump of a greenhouse and then send the information obtained by the sensors by SMS
and perform operations with a smartphone. In [20] an irrigation system is proposed that
allows the optimization of water, using humidity, temperature sensors and a flow sensor
in the field as inputs for a fuzzy Mamdani controller, while the outputs are the ignition
of a lamp and an irrigation pump. Finally, in [21] an intelligent irrigation system is
presented that uses an Android application to optimize water consumption in medium-
scale gardens and fields using sensors that capture plant data and weather factors in real
time. Fuzzy logic is used for decision making and activating the irrigation pump.
This paper presents the design and development of a controller for an automated
irrigation system using fuzzy logic. The controller developed in this project uses
sensors to measure soil moisture, ambient humidity, luminosity and ambient temper-
ature to determine the activation of an irrigation pump. The soil moisture is periodically
monitored to determine if the humidity needed to measure the other sensors is met and,
if so, the data obtained from the sensors feeds a Sugeno FIS based on specific rules for
habanero pepper plants; the system output activates or deactivates a pump for drip
irrigation in a greenhouse.
2 System Design
The development of the controller consists of two subsystems, (1) detection and control
by means of the sensors, and (2) the internal development of the controller. The
development of these subsystems is described in more detail in the following
subsections.
2.1 Sensing and Control

For the operation of the controller, three specific sensors are required: soil moisture
sensor (FC 28), light sensor (LDR) and humidity and temperature sensor in the
environment (DTH 11), which are connected to a Raspberry module, an electrovalve
and finally to a power supply. The sensors perform readings that are sent to the
controller to process the values of each of them and based on the fuzzy rules the
activation of the solenoid valve is determined [22].
The operation of the controller is observed in Fig. 1. The controller performs a
reading of the sensors every 10 min; the values thrown are verified by means of the
fuzzy inference mechanism to determine if the rules to activate the irrigation system are
met; once irrigation has started, a reading of the values of the sensors every 2 min is
made to determine the moment in which the irrigation must be stopped.
Fig. 1. Operation of the fuzzy controller.

A conventional solenoid valve is used to supply the water, since the irrigation
system works only in two states: on and off. The solenoid valve is connected to a hose
that works in conjunction with sprinklers and is activated by the fuzzy controller when
complying with the rules presented in Table 1. These rules were defined by interviews
with experts in the care of habanero pepper when presented with different scenarios
with respect to the input variables. Soil moisture sensors are the ones that mainly
determine if the activation of the system is necessary, because when the soil is com-
pletely humid, irrigation is not carried out regardless of the values of the other sensors,
however, if the soil moisture sensors are in a “not wet” range (as described below), then
the values of the other sensors are considered to determine the activation of the sole-
noid valve with the calculation of the rules.
Table 1. Rules to determine whether the irrigation pump should be turned on or off, according
to experts in the care of habanero pepper.
Soil moisture Ambient temperature Ambient humidity Luminosity Irrigation pump state
High – – – Off
Medium Low High – Off
Medium Low Medium Not low Off
Medium Low Medium Low On
Medium Low Low Not low Off
Medium Low Low Low On
Medium Medium – Not low Off
Medium Medium – Low On
Medium High – Not low Off
Medium High – Low On
Low Low High – On
Low Low Medium – On
Low Low Low Not high On
Low Low Low High Off
Low Medium – Not high On
Low Medium – High Off
Low High – Not high On
Low High – High Off
2.2 Fuzzy Logic Control

The scheme of the fuzzy controller can be seen in Fig. 2. A Sugeno FIS is commonly
used when discrete outputs are required because the membership functions associated
with the output variable are linear or constant. Because the pump that controls the
irrigation system has only two states, on and off, a FIS based on the zero-order Sugeno
model is the appropriate choice. As a first step, the controller receives as input data the
readings of the moisture sensors in the soil, s1; humidity of the environment, s2;
luminosity, s3; and temperature of the environment, s4. The variables are evaluated
(fuzzificated) according to the different rules on sigmoidal and Gaussian membership
functions, whose parameters were defined empirically and, in the same way as the
fuzzy rules, by means of interviews with experts in the care of habanero peppers. These
values represent the input of the FIS, whose output determines the switching off or on
of the irrigation pump. All the membership functions of the output variables in a
Sugeno FIS are linear, and in our case, constants of the form z = c, where z represents
the output variable and c a scalar value, in our case 0 (off) or 1 (on).
Fig. 2. Fuzzy controller scheme.
The membership functions are distributed according to the linguistic values

established for each linguistic variable. Subsequently, the values are evaluated with the
fuzzy inference mechanism, that is, the rules established in the base of rules are
applied; the decision taken in each situation must take into account the fuzzification
process, the result of the mechanism of inference with the minimum and maximum
degree of belonging and the Sugeno base of rules. Finally, defuzzification is applied by
calculating the weighted average so that the controller produces the desired output. The
membership functions of the entries are shown in Fig. 3.
The operation of the system is briefly described below. For each rule, the data
obtained from the sensors s1, s2, s3 and s4, are evaluated in their corresponding
membership functions f1, f2, f3 and f4, and these results are used to calculate the firing
strength of each rule. Because all the rules described in Table 1 use the logical con-
junction (AND) and a Sugeno model is being used, the firing strength of rule i is
calculated as
wi ¼ f1 ðs1 Þf2 ðs2 Þf3 ðs3 Þf4 ðs4 Þ: ð1Þ
The final output Z of the system is calculated (defuzzified) by means of the

weighted average (based on the different firing strengths) of all the outputs for each
rule, calculated as:
P
18
wi zi
Z ¼ i¼1 : ð2Þ
P18
wi
i¼1
Fig. 3. The membership functions used in the FIS, (a) those for soil moisture, (b) luminosity,
(c) ambient humidity and (d) ambient temperature.
A value of Z close to 1 turns on the irrigation system (or keeps it on, according to
the situation), while a value close to 0 turns it off (or keeps it off).
3 Results
According to the input values of each sensor, the changes in the readings of the values
obtained can be observed, therefore, compliance with the rules is evaluated to deter-
mine whether the solenoid valve should be turned on or off. Figure 4 graphically shows
the operation of the system. Each row represents each of the 18 rules that regulate the
behavior of the controller. The first four columns represent the different variables
obtained from the sensor data: soil moisture, luminosity, ambient temperature and
ambient humidity. The last column represents the activation (or not) of the irrigation
pump based on each of the rules. The graph in the lower right corner represents the
aggregation of the different outputs for each rule; it is this result that determines
whether the irrigation pump is to be turned on or off.
Fig. 4. Result of the operation of the FIS. Each row represents each of the 18 rules that regulate
the behavior of the controller. The first four columns represent the different variables obtained
from the sensor data. The last column represents the activation (or not) of the irrigation pump
based on each of the rules. The graph in the lower right corner represents the aggregation of the
different outputs for each rule. The graph shows a case of 35.3% moisture in the soil, 72.2%
luminosity, 26.7 °C ambient temperature and 78.4% ambient humidity; the irrigation pump
would be turned off in this case due to an aggregated value close to 0.
The operation of the fuzzy controller turns out to be as expected, since the irrigation
system stays off when the soil is totally humid and there are high temperatures. In
addition, the controller correctly complies with readings of the values of the sensors
every 10 min when the pump is off and readings every 2 min when it is turned on.
4 Conclusions
In this work we describe the design of a fuzzy controller based on the Sugeno model for
the control of an automatic irrigation system specifically designed for habanero pepper.
Based on interviews with experts in the care of habanero pepper, 18 if-then fuzzy rules
were established and the different membership functions corresponding to linguistic
values of the four linguistic variables obtained from different sensors were designed:
soil moisture, luminosity, ambient temperature and ambient humidity. These mem-
bership functions were designed with Gaussian and sigmoidal functions.
The implementation of fuzzy logic in a controller for an irrigation system implies a
great impact and benefit for agriculture and the people who dedicate themselves to this
activity, due to the automation of tasks that allows people to reduce time and processes
that would normally need to be carried out by more than one person. It allows pre-
serving the care of the water by having an automatic on/off control based on weather
factors, in order to care for crops and thus help the farmer to increase productivity. The
design of an irrigation system like the one proposed in this paper can also be extended
to the specific conditions of plants other than the habanero pepper plant by redesigning
the fuzzy rules and associated linguistic values.
Acknowledgements. The authors want to acknowledge the Tecnológico Nacional de México/I.T.

Mérida for the financial support provided through projects 6449.19-P and 6714.19-P.
References
1. Traxco: Historia ancestral del riego agrícola - El agua de riego. https://www.traxco.es/blog/
noticias-agricolas/historia-ancestral-del-riego
2. World Water Council: Visión Mundial del Agua-Resumen Ejecutivo. http://www.
worldwatercouncil.org/fileadmin/wwc/Library/Publications_and_reports/Visions/Spanish
ExSum.pdf
3. Ruiz, F., Esquivel, K., Rodríguez, D., Rodríguez, M., Duarte, R.: Internet of things (IoT), an
alternative for the care of water, vol. 40, pp. 2318–2330 (2018)
4. Calzada, J., Narváez, J., Aguilar, R., Velasco, H.: Agenda Técnica Agrícola Yucatán,
México (2015)
5. Narváez, M.: Chile habanero de Yucatán, único en el mundo. http://www.cienciamx.com/
index.php/ciencia/mundo-vivo/22848-chile-habanero-yucatan
6. Andrade Rodríguez, M., Villegas Torres, O.G., Juárez López, P., López Gómez, J.D.,
Martínez Fernández, E., Sotelo Nava, H.: Rendimiento y calidad del chile habanero
(Capsicum chinense Jacq.) por efecto del régimen nutrimental. Rev. Mex. Ciencias
Agrícolas 8, 1717 (2018). https://doi.org/10.29312/remexca.v8i8.699
7. Tun, J.: Chile habanero características y tecnología de producción. Secr. Agric. Ganad.
Desarro. Rural. Pesca y Aliment. 79 (2001)
8. Ruiz, N., Medina, F., Martínez, M.: El chile habanero: su origen y usos (2011)
9. Scott, C., Banister, J.: The dilemma of water management “regionalization” in México under
centralized resource allocation. Int. J. Water (2008). https://doi.org/10.1080/0790062070172
3083
10. Baum, B.A., Titlow, J., Vasanth, T., Welch, R.M.: Automated cloud classification of global
AVHRR data using a fuzzy logic approach (1997)
11. Jang, J.-S.R., Sun, C.-T., Mizutani, E.: Neuro-Fuzzy and Soft Computing Jang: A
Computational Approach to Learning and Machine Intelligence. Prentice-Hall Inc., Upper
Saddle River (1997)
12. Zadeh, L.: Fuzzy sets, Berkeley, California (2013). https://doi.org/10.1007/978-1-4614-
1800-9_73
13. Ying, H.: Fuzzy control and modeling : analytical foundations and applications. IEEE Press
Series in Biomedical Engineering (2000)
14. Ying, H.: The simplest fuzzy controllers using different inference methods are different
nonlinear proportional-integral controllers with variable gains. Automatica 29, 1579–1589
(1993). https://doi.org/10.1016/0005-1098(93)90025-O
15. Touati, F., Al-Hitmi, M., Benhmed, K., Tabish, R.: A fuzzy logic based irrigation system
enhanced with wireless data logging applied to the state of Qatar. Comput. Electron. Agric.
98, 233–241 (2013). https://doi.org/10.1016/j.compag.2013.08.018
16. Hussain, M.H., Min, T.W., Siraj, S.F., Rahim, S.R.A., Hashim, N., Sulaiman, M.H.: Fuzzy
logic controller for automation of greenhouse irrigation system (2011)
17. Suruthi, N., Saranya, R., Subashini, S., Shanthi, P., Umamakeswari, A.: Managing irrigation
in Indian agriculture using fuzzy logic – a decision support system. Int. J. Eng. Technol. 7,
321 (2018). https://doi.org/10.14419/ijet.v7i2.24.12075
18. Hilali, A., Alami, H., Rahali, A.: Control based on the temperature and moisture, using the
fuzzy logic. Int. J. Eng. Res. Appl. 07, 60–64 (2017). https://doi.org/10.9790/9622-
0705036064
19. Parmar, B., Chokhalia, J., Desarda, S.: Terrace garden monitoring system using wireless
sensor networks. Eng. Technol. Manag. 02, 51–54 (2019)
20. Izzuddin, T., Johari, M., Rashid, M., Jali, M.: Smart irrigation using fuzzy logic method.
ARPN J. Eng. Appl. Sci. 13, 517–522 (2018)
21. Munir, M., Bajwa, I., Cheema, S.: An intelligent and secure smart watering system using
fuzzy logic and blockchain. Comput. Electr. Eng. 77, 109–119 (2019). https://doi.org/10.
1016/J.COMPELECENG.2019.05.006
22. Boman, B., Smith, S., Tullos, B.: Control and Automation in Citrus Microirrigation Systems,
Florida (2008)
Internet of Things for Irrigation System
J. M. Guzmán-Toloza, D. F. Villafaña-Gamboa(&),
L. J. Peniche-Ruiz, R. A. Gómez-Buenfil,
J. K. Molina-Puc, and M. J. Rodríguez-Morayta
Departamento de Sistemas y Computación, Tecnológico Nacional de México,

I. T. de Mérida, Av. Tecnológico km. 4.5 S/N, 97118 Mérida, Yucatán, Mexico
{e15080844,dakar.villafana,larissa.peniche,e15080988,
e15081607,manuel.morayta}@itmerida.edu.mx
Abstract. Technological advances have impacted in recent years most of the

economic sectors of the world, where they have managed to increase efficiency
and profitability particularly with one of the fastest growing applications in this
sector: the Internet of Things (IoT). Agriculture is not the exception. However,
most farmers in Mexico still use traditional methods of control and monitoring,
which prove to be inefficient, causing time and cost losses in today’s globalized
world.
The present design is an intelligent irrigation system, implemented under IoT
principles and free hardware and software. There are soil moisture, luminosity,
humidity and temperature sensors in the environment connected to a Raspberry
Pi module; sensors detect values in real time while the Raspberry sends the data
to be stored in the cloud. The control system is based on fuzzy logic rules that
allow turning on and off a water pump. This design includes a mobile app where
the system status is monitored. Tests show an appropriate behavior of the irri-
gation system.
Keywords: App mobile application Internet of Things Raspberry Pi

Irrigation system Sensors
1 Introduction
In the last 100 years, the world’s population has tripled, while water withdrawals have
increased six fold, increasing the pressure on natural resources and specifically on
water resources worldwide [1].
Agriculture is one of the main pillars of the Mexican economy, 76% of all available
water is destined for agricultural use, while for public supply only 14.4%. In Mexico,
most irrigation systems operate manually, which in practice ends up being of little use,
promotes excessive use and even empirical use of water by not having a means to
verify whether the plants have had optimal watering. Among the main difficulties for
the technification of irrigation stand out: the high costs of conditioning the irrigation
infrastructure, low efficiency of conduction and distribution due to filtration problems,
excessive parceling of agricultural land, low efficiency of water application, due to
ignorance of efficient irrigation methods and the requirement of water per crop; the
weak organization of farmers and their low participation, so that innovation in irrigation

https://doi.org/10.1007/978-3-030-33229-7_25
Internet of Things for Irrigation System 295
technology has become a priority in modern times, in order to make agricultural

production more efficient and, at the same time, save the vital liquid [2].
Irrigation systems are composed of different elements such as hoses or pipes, as the
case may be; step valves and motor pumps, which allow the flow of water through the
system to the crop or land to be irrigated [3]. In recent years, technologies have been
developed to automate or semi-automate irrigation, these technologies have come to
replace the traditional agricultural irrigation mechanism [4]. At the end of the 90 s, the
Internet was mainly used as a tool to search for information. In the last 10 years there
has been a new way of using the Internet, where everything has become social,
transactional and mobile [5]. Currently, society is immersed in what can be said in the
second generation of the Internet, called IoT [6]. It is known as IoT to the ability to
create connection between objects of the everyday world, allowing to share a lot of
information and store it in a database that in most cases is in the cloud [7], thus having
control and access to it information from wherever a person is, obviously with internet
access [8].
A field of action of the IoT, is its application is the agricultural sector [9], thus
creating precision agriculture, which consists in applying the tools offered today by
information technology and communications together with different sensors and tools
of IoT, in the administration, management of the crops and their land, taking into
account the variables and properties that are important in order to improve crop yield,
profitability and environmental quality.
Technology can be applied to monitor important characteristics of the plants that
the farmers require to keep the crops in good growth condition. Agriculture is not an
exact science and it depends on experience-based values; atmospheric conditions of
plants in an outdoor field vary from place to place, which makes it difficult to maintain
uniformity in all areas of the field [10]. This situation can be better approached from a
fuzzy logic point of view, as it has been implemented in various investigations for both
fields and greenhouses with efficient irrigation results on different crops.
It can be said that the technological universities of Ecuador, Colombia and Spain to
name a few, have made similar designs where they use the Arduino card as the actuator
node and the Raspberry module as the coordinating node or gateway, and they also use
an app for system control. However, they still do not use fuzzy logic as a control
system, which is an innovative branch of artificial intelligence for the present design.
This paper presents an intelligent irrigation system based on IoT [11] with an
integrated system to monitor values in real-time using a Raspberry Pi microcontroller
together with a number of sensors. The Raspberry Pi works as a gateway between the
information collected by the sensors and a real-time database that constantly stores and
updates the values of all sensors. It also has a mobile application through which it is
possible to verify the state of the system, the eventualities and the values of the sensors
in real time. This system offers an easy to implement wireless solution with water
savings using technification as opposed to traditional irrigation systems.
296 J. M. Guzmán-Toloza et al.
2 System Architecture
The system architecture consists of 3 modules shown in Fig. 1: Sensors module,

Embedded system module and Mobile application module. The following section
contains a detailed description for each one.
Fig. 1. Architecture for IoT-based monitoring.
2.1 Sensors Module

There are three sensors used according to the parameters: Luminosity, soil humidity
and environmental humidity and temperature.
LDR. This sensor measures the luminosity in the environment. It is a resistor that
varies its electrical resistance value depending on the amount of light that falls on it, it
is also called photo resistor. The electrical resistance value is low when there is light
affecting it and very high when it is in the dark. The variation of its resistive value has a
certain delay, which is different if it goes from dark to lighted or from lighted to dark.
YL-69. This sensor was used to measure the humidity in the soil as it has the ability to
measure soil moisture. It works applying a small voltage between the terminals of the
module, a current passes depending on the resistance that is generated in the soil and it
depends a lot on moisture. Therefore, when the humidity increases the current increases
and when the current decreases it decreases [12].
DHT11. This sensor was used to measure the humidity and temperature in the envi-
ronment. It is a digital sensor of temperature and relative humidity in the environment.
It integrates a capacitive humidity sensor and a thermistor to measure the surrounding
air; it displays the data by means of a digital signal on the data pin. This sensor is
calibrated at the factory to obtain calibration coefficient recorded in its OTP memory,
ensuring high stability and reliability over time. The communication protocol between
the sensor and the microcontroller uses a single wire or cable [13].
2.2 Embedded System Module

This module is developed in the integrated Raspberry Pi 3 Model B system, which has
the following main features: Cortex A53 quad-core with cache of 512 Kbyte L2
dedicated and 1.2 GHz, 1 GB of RAM, BCM43438 wireless LAN and operating
system based on Linux, the kernel version used is 4.14 [14]. The integrated system has
two servers: the data server and the IoT server, implemented in an integrated system
based on a Linux distribution for Raspberry known as Raspbian. The data server is
responsible for handling the requests and receiving the corresponding information from
all the sensors that are being monitored. The IoT server is responsible for receiving the
requests of the database in real time, sending the stored information of the sensors via
Wi-Fi and storing it in the cloud. From the mobile application the corresponding
requests are sent to get the updated information to be displayed on the device’s screen.
2.3 Mobile Application Module

The mobile app allows displaying the real-time values collected on the database by
means of the sensors constantly registering data. This application, developed on
Android operating system with the Java programming language, can run on any mobile
device or tablet with Android 5.0 Lollipop version onwards. Figure 2 shows the case of
use with 3 main functions of the mobile application.
Fig. 2. Case of use of the mobile application.
The first two cases verify irrigation history and check sensor status; they provide a
broad overview of the functionality of the system: the first one shows the user the exact
date and time when irrigation started while the second shows the status of each sensor;
this is useful to identify problems in the data collection or any malfunction. The last
function, check sensor values, allows real time visualization of the values obtained by
the installed sensors.
The V methodology was applied to develop the system [15], where tests for each
module such as the data server and the IoT server, are first carried out. Next, hardware
and software are integrated to verify the correct operation of both systems as a whole.
Finally, the functionality tests of the system in general are carried out.
2.4 Sensors Tests

To verify the correct operation of each sensor (DHT11, LDR y YL-69), they were
connected and tested one by one (see Fig. 3), until obtaining the data registered by the
sensors for the server to process the information. The request is made from the
Raspberry terminal by means of a Python code that lists the most recent information
obtained by each sensor. To avoid information overlapping and to assure reliable
information, a delay of 2 s between readings of one sensor and the next.
Fig. 3. Physical connections of the sensors.
Figure 3 shows the pins through which each sensor sends the recorded readings;
each pin is connected in series with a resistance of 220 X to limit the current and thus
avoid burning the Raspberry input terminal. The Raspberry Pi 3 module provides each
sensor with a +5 Vdc supply to operate. The GPIO Raspberry terminals of the irrigation
system electronic circuit are illustrated in more detail in the interconnection diagram in
Fig. 4.
Fig. 4. Schematic diagram of the connections.
Once the information server has collected current data from all sensors, these are
displayed in Raspberry terminal, so it is easy to visually verify that both the server and
the sensors are working correctly, thus there is communication between them without
interruptions or loss of information. Table 1 illustrates an example of the detected
values.
Table 1. Example of sensor values in the CMD terminal.

Time Environment Humidity in the Soil Brightness
delay temperature environment moisture
2 seg 33 °C 80% 30% 34%
4 seg 33 °C 79% 32% 35%
6 seg 33 °C 78% 31% 33%
8 seg 34 °C 79% 28% 35%
10 seg 34 °C 80% 29% 36%
12 seg 34 °C 79% 30% 37%
The information server sends the current data to the IoT server, which has the
necessary configurations to make the connection through the internet with the database
in real time.
In this project, the design and development of a controller for an automated irri-
gation system using fuzzy logic is carried out. The developed controller uses sensors
for soil moisture, humidity in the environment, brightness and temperature in the
environment to manage the activation of an irrigation pump; the moisture in the soil is
monitored every 10 min to determine if the humidity necessary to read the other
sensors is met and if each of the necessary factors is met, the irrigation system is
activated.
A conventional solenoid valve is used to supply the water, since the irrigation
system works only in two states: on and off. The solenoid valve is connected to a hose
that works in conjunction with sprinklers and it is activated by the fuzzy controller
when complying with the rules presented in Table 2. These rules were defined based on
interviews with experts in the care of habanero pepper when presented with different
scenarios with respect to the input variables. Soil moisture sensors are the ones that
mainly determine if the activation of the system is necessary, because when the soil is
completely humid, irrigation is not carried out regardless of the values of the other
sensors.
Table 2. Rules to operate the irrigation system.

Soil Environment Humidity in the Brightness Irrigation pump
moisture temperature environment state
High – – – Off
Medium Low High – Off
Medium Low Medium Not low Off
Medium Low Medium Low On
Medium Low Low Not low Off
Medium Low Low Low On
Medium Medium – Not low Off
Medium Medium – Low On
Medium High – Not low Off
Medium High – Low On
Low Low High – On
Low Low Medium – On
Low Low Low Not high On
Low Low Low High Off
Low Medium – Not high On
Low Medium – High Off
Low High – Not high On
Low High – High Off
2.5 Tests of the IoT Server

The IoT server (located in the Raspberry) waits for a request to obtain real time
information from the database (located in the cloud), which in turn receives another
similar request from the client of the mobile application module (located on the
Smartphone) to obtain the latest information from the sensor module (located near the
crop plants), represented as a block diagram in Fig. 1.
The IoT server has a established project address in the database, and the infor-
mation destination nodes connected to it in order to send each value to the corre-
sponding node. This ensures that the sensed data reaches the correct destination. Once
the IoT server accepts a request, the data server makes a request for information to the
sensors and then passes it to the IoT server to send it to the database in real time with an
established format. The mobile application for its part makes a request to the database
in real time to obtain the current information in the established nodes and then
immediately displays it in the user interface for viewing (see Fig. 5).
a. Main view b. Sensors status view c. Irrigation history view
Fig. 5. Views of the mobile application.
The mobile application allows representing the values obtained from each of the
sensors through graphs due to the integration of a Microsoft Office Excel® plugin,
which facilitates the analysis and understanding of the behavior of the data. Two graphs
are shown in Figs. 6 and 7, corresponding to average values obtained from the lumi-
nescence and humidity sensors in the soil.
Fig. 6. Average luminosity graph.
Fig. 7. Average humidity in the environment graph.
The graphs allow seeing the behavior of the sensors and facilitate the decision
making to perform the irrigation.
3 Conclusions
There are already autonomous and automated irrigation systems in the market, but in
most cases their costs are too high for the common farmer, making them inaccessible.
This is why the proposed system considerably reduces the cost to make it available to
rural people, who in most cases lack economic resources. Its operation is as simple as
knowing the basics to manage applications and a smartphone or tablet with internet
access. Furthermore, if we consider that the situation of water scarcity will worsen this
kind of projects for the agricultural sector may contribute to solve this global problem
and keep the vital liquid in optimal usage.
The architecture that makes up this project allows monitoring of parameters in the
irrigation system, using a data server and an IoT server implemented in an integrated
system based on a distribution of Linux for Raspberry known as Raspbian. The data
server receives the information from the sensors every 2 s to avoid data overlapping
and information loss. Therefore, the proposed architecture can perform real-time
monitoring at low cost on the irrigation system. The IoT server sends the information of
the sensors to a database in real time, which in turn provides the updated information to
a mobile application implemented on a smartphone or tablet.
Implementing fuzzy logic in a controller for an irrigation system implies a great
impact and benefit for agriculture and people who are engaged in such activity, due to
the automation of tasks that allow people to reduce time and processes, based on
weather factors, using water in an efficient way avoiding waste and providing the plants
with sufficient quantity for optimum growth.
On a first stage, the irrigation system was developed and tested at laboratory level;
it will subsequently be installed in a greenhouse to grow habanero pepper plants. In the
future, more sections would be implemented within the mobile application, to have a
greater amount of information for the irrigation system, with the advantage to be able to
monitor real time data for better crops administration. It is also intended to integrate a
digital image processing system with a neural network, for the prediction of crop pests
and registration of plant growth, thus increasing the benefits of the intelligent irrigation
system.
References
1. Comisión Nacional del Agua: Estadísticas del Agua en México, pp. 45–50 (2018)
2. Fereres, E., García-Vila, M.: Irrigation management for efficient crop production. In: Savin,
R., Slafer, G. (eds.) Crop Science. ESSTS, pp. 345–360. Springer, New York (2019). https://
doi.org/10.1007/978-1-4939-8621-7_162. https://link.springer.com/referenceworkentry/10.
1007%2F978-1-4939-2493-6_162-3
3. IICA: Innovation and Water Management for Sustainable Development in Agriculture,
pp. 20–29 (2015). http://repositorio.iica.int/bitstream/11324/3035/2/BVE17068948i.pdf
4. Organización de las Naciones Unidas para la Alimentación y la Agricultura: El estado
mundial de la agricultura y la alimentación. Roma (2018)
5. Pastor, E.J.: Introducción al Internet de las Cosas, pp. 80–120 (2015)
6. Clúster ICT-Audiovisual de Madrid: Internet de las cosas: Objetos interconectados y
dispositivos inteligentes, pp. 68–75 (2016). https://actualidad.madridnetwork.org/
imgArticulos/Documentos/635294387380363206.pdf
7. Negrete, J.: Internet of things in Mexican agriculture; a technology to increase agricultural
productivity and reduce rural poverty, pp. 46–47 (2018)
8. Alper Akkas, M.: An IoT-based greenhouse monitoring system with Micaz motes, pp. 604–
608 (2017). https://www.sciencedirect.com/science/article/pii/S187705091731709X
9. World Water Vision Report: Visión Mundial del Agua, México (2017)
10. Karim, F., Karim, F.: Monitoring system using web of things in precision agriculture. In:
The 12th International Conference on Future Networks and Communications (2017). https://
reader.elsevier.com/reader/sd/pii/S1877050917312590
11. Shekhar, Y., Dagur, E., Mishra, S., Tom, R.J., Veeramanikandan, M., Sankaranarayanan, S.:
Intelligent IoT based automated irrigation system. Int. J. Appl. Eng. Res. 12(18), 7306–7320
(2017)
12. OBSoil-01: Octopus Soil Moisture Sensor Brick, p. 5 (2016)
13. D-Robotics: DHT11 Humidity & Temperature Sensor, Mouser Electronics. https://www.
mouser.com/ds/2/758/DHT11-Technical-DataSheet-Translated-Version-1143054.pdf.
Accessed 27 May 2019
14. Raspberry Pi Foundation: Raspberry Pi 3 Model B, Raspberry Pi Website, p. 137 (2016)
15. Perez, A., et al.: Una Metodología Para El Desarrollo De Hardware Y Software Embebidos
En Sistemas Críticos De Seguridad. Sist. Cibernética e Informática 3(2), 70–75 (2014)
Architecture Proposal for Low-Cost
Hydroponic IoT Based Systems
Gutiérrez León Evelyn1(&), Jorge Erik Montiel1,

Carreto Arellano Chadwick2, and Felipe Rolando Menchaca García1
1
Escuela Superior de Ingeniería Mecánica y Eléctrica unidad Zacatenco,
evelyngleon@hotmail.com, Jerikmontiel.isc@gmail.com,
fmenchac@gmail.com
2
Escuela Superior de Cómputo, Instituto Politécnico Nacional,
Mexico City, Mexico
ccarretoa@ipn.mx
Abstract. This article proposes the design of an intelligent management system

based on Internet of Things (IoT) for hydroponic crops, to bring farmers and
marginalized sectors closer to the use of new technologies, increasing precision
in agriculture according to the needs of the crop in “real time.” On this work, we
propose the development of prototypes based on the architecture and network
topologies for IoT devices, as well as the technologies available in the market.
All this, to acquire information about the environment of the crop and to help the
user on the decision making. The premise of the project is to develop a modular
solution, scalable, low cost, easy installation, management, and operation, to
potentiate the technological update of the systems of crops already installed or
projects to start.
Keywords: Internet of Things Precision agriculture Hydroponics systems

Smart farming
1 Introduction
Agriculture is the third economic activity in the country, about 22 million hectares are
dedicated to the agriculture, of which, 75.67% implement some level of mechanization,
and 64.92% of this is seasonal crop [1].
Of the agronomic population, approximately 35% of the farmers are in the micro, or
small business sector with ties in the domestic market and 50% produce for self-
consumption. Despite the existence of diverse imported solutions for the agro-food
field, these are focused on intensive producers of the industrial scale. They require high
initial investment and qualified personnel in the technology for its installation, oper-
ation, and maintenance. For this reason, its implementation has not been generalized
for producers in the micro and small enterprise sectors as well as self-consumption
orchards, who are more vulnerable to climate variability and change [2], generating a
technological and economic gap between small farmers versus intensive producers.

https://doi.org/10.1007/978-3-030-33229-7_26
306 G. L. Evelyn et al.
Also, with the growing urban expansion, areas dedicated to agriculture have been
reduced or displaced to forests and jungles, irremediably damaging the ecosystem by
decreasing the participation of the population in agriculture.
According to this, the design of low-cost solutions based on alternative systems that
potentiate the results of the agricultural campaign in small spaces, such as hydroponics,
along with precision agriculture and the Internet of Things (IoT), allow the modern-
ization of agriculture in the mentioned sectors without huge investments, obtaining
better quality products, this with the aim of generating economic growth in the region.
1.1 Hydroponics Systems Characteristics

Hydroponics is the technique of production or cultivation without soil, in which water
and nutrients are supplied to the plants through a complete nutritious solution (NS),
providing the necessary conditions for its proper development [3].
Hydroponic systems are characterized by how the nutrient solution for the crop, the
type of module, and the arrangement of the elements, as well as the size, space, and
environmental conditions, are available. Therefore, a hydroponic system can vary from
one farmer to another, or even within the same set of crops. Therefore, the design of the
solution is focused on a series of devices to be placed in the hydroponic modules,
regardless of their position, size or location, or if they are commercial or home
modules.
1.2 Precision Agriculture

The Precision Agriculture (PA) also known as “Smart Farming” in the words of
Patokar and Gohokar is “the technique of applying the appropriate amount of input
(water, pesticides and fertilizers, among others) at the right time and place to improve
production and increase crop yield” [4]. Its central premise is to obtain as much
information as possible from the environment and the crop (in order to predict the
changes and needs in the different stages of growth) to provide the stimuli and nutrients
necessary for the proper development of the plant. where the use of IoT has been
widely implemented [5]. The critical parameters of the hydroponics crops are the light
intensity, ambient temperature and humidity, and the temperature of the NS [6, 7] cause
affect the NS absorption directly. For the irrigation system, we are focused on the
irrigation period and duration.
1.3 Internet of Things in the Agriculture

The use of networks of objects equipped with sensors (also called wireless sensor
networks or WSN) has expanded in the agricultural field, due to the efficiency it has
shown in obtaining information from it. Therefore, its study and implementation have
increased in solutions based on the IoT communication paradigm [8].
The IEEE-SA Working Group has a project to develop an IoT standard: IEEE
P2413. On the IEEE P2413 define an IoT architecture framework for the architectural
needs of different IoT scenarios (see Fig. 1) [9].
Architecture Proposal for Low-Cost Hydroponic IoT Based Systems 307
Fig. 1. Three-layer architecture for IoT proposed by the IEEE Working Group P2413.
The IoT solutions have many challenges to be facing [10–12]: Scalability, inter-
operability, mobility and adaptable solutions in order to provide evolved services. The
edge computing and Fog architectures helps to stablish the IoT architectural definition
and support its applications [13, 14] trough the implementation of different security
mechanism stablished for Internet protocols and smart objects standards.
The first stage for the IoT solutions is about obtain all the behavior information of
the study case, then analyze all the data collected in order to propose the algorithms for
the decision making and finally the final solution needs to be tested and corrected to
improve the results of the solution trough some advance control algorithms.
This article describes an architectural proposal for the development of a hydroponic
management system using IoT technologies. In the following section, the analysis and
design of the system are addressed, focusing on the development of two different
networks, the intra-greenhouse network and the extra-greenhouse network which, in
conjunction with sensors and actuators, monitors and control the greenhouse operation
variables.
The third section shows the development of communication networks as well as the
development of specific frames within the network, which allow optimizing the
implemented communication. Subsequently, the implementation and the obtained
results of the development of the system are briefly explained, and finally, we conclude
with possible future improvements of the system.
2 Design
Between crop campaigns, the distribution and extension of crops is variable, for this
reason, it´s important design a network topology of the solution with the follow
characteristics: modular in order to cover the needs of crop management at different
scales; flexible and dynamic to allow variations in the geographical arrangement and
configuration of the nodes; low power to offer the autonomy of the nodes; and tolerant
to failures to ensure the operation of the network. According to this, the solution was
divided into two action zones:
Intra-Greenhouse. A wireless network is proposed to facilitate the installation and
communication of the modules without tools or extra material, the network is based on
a mesh topology to meet the requirements of a robust and dynamic communication
with enough coverage between the nodes arranged inside the greenhouse. Within this
network can define different types of nodes which focus on different roles within the
greenhouse, at least 3 types of nodes are proposed: Irrigation Node, Sensor Node and
Master Node. In the following sections will detail each of them.
Extra-Greenhouse. It’s required the transmission of data collected inside the green-
house to the server for further processing. The solution considers the communication of
the server to one or more greenhouses, so the network topology is defined as a star
topology in this communication stage. A wireless network is proposed to facilitate the
installation and communication of the modules without tools or extra material, the use
of a mesh topology could satisfy the requirements of a robust and dynamic commu-
nication with enough coverage among the nodes arranged inside the greenhouse.
Subsequently, the mobile terminal requires obtaining the data stored on the server.
To do so, it is necessary to consider the following cases:
Local network connection: The server will serve as a wireless access point to
download the information collected in the mobile terminal.
Anywhere connection through the Internet: To achieve this, it is necessary to
contract a service in the cloud for the exchange of data from the server to the mobile
terminal through the Internet and the necessary infrastructure exists.
Finally, we can define a hybrid network architecture integrated by one or more
“intra-greenhouse” networks which operate with a general “extra-greenhouse” network.
These networks can be of different protocols if there is a node that works as middleware
between both networks.
2.1 Network Architecture Model

The minimum case of the proposed architecture is observed when only one greenhouse
is available (as in subsistence gardens, urban or backyard gardens), that is, where the
coverage area is smaller. The Fig. 2 shows the use of sensor nodes with the function of
capturing information about the crop environment; an irrigation node, which will
collect information and control the status of the irrigation system, and a master node in
charge of coordinating the network and concentrating the information provided by
other nodes, repackaging the data and communicating them to the server through the
Extra-network. Greenhouse. Subsequently, the server processes and stores the infor-
mation to be displayed to the user through a graphical application interface for mobile
devices.
This architecture can be replicated again based on the number of greenhouses that
you want to manage within a location. This second case can be seen in the Fig. 3,
Fig. 2. The general architecture of the proposed network for the intelligent management system
for hydroponic crops, minimum case.
Fig. 3. Extra-greenhouse network diagram for multiple greenhouses
where we have multiple greenhouses, each one with its Intra-Greenhouse network
connected to the same Extra-Greenhouse network and with it to the same server.
This modularity can be applied in the same way at the greenhouse level, so that we
can have multiple greenhouses with different needs, in this case, it’s the Intra-
Greenhouse network that varies in the number of nodes, adjusting to the characteristics
of each place, this can be visualized in the Fig. 4.
Fig. 4. Intra-Greenhouse network diagram: (a) minimum case, (b) expansion of the network
after adding sensor nodes.
2.2 Wireless Sensor Network

Inside each greenhouse, three types of nodes are deployed that which works with
specific tasks of capturing the information and activation of actuators depending on
their role. The design of the nodes is made according to the coverage area, low cost,
easy installation, spatial and connection independence, and compatibility between
components. The block diagram of each sensor is specified below.
Sensor Node. These nodes are designed to obtain the current state of the environment
in the hydroponic module. The block design of the sensor node is defined in 4 main
blocks: Perception, Feeding, Processing and Control, and the communication block.
The unit of perception in the sensor node has the necessary transducers to capture the
information of the environment: Lighting (lx), Temperature (°C), Humidity (%RH) and
Temperature of the nutrient solution (°C). These transducers are connected to the
microcontroller unit through the I2C, UART and SPI interfaces, as well as the GPIO,
while the wireless transceiver nRF24L01+ communicates through the inter - SPI of the
microcontroller. By the characteristics of the previously mentioned requirements, the
AtMega328p microcontroller was selected as the MCU of the sensor node. The power
supply is made up by a Li-Po battery and a charger module connected to the main
power.
Sampling time of 15 min was established to collect the information from the
environment. Once the sample is taken, the data is pre-processed and packaged by the
MCU to form the defined frame. Then the transmitter mode of the transceiver is
enabled, of and the MCU sends the frame the to the master node directly or by the
nearest node through the forwarding of packets. The transceiver changes to receiver
mode and after a listening period is set to wait until it is activated again in the next
reading of the environment (see the Fig. 5).
Irrigation Node. This type of node is responsible for obtaining information on the
status of the nutrient solution through the pH and Temperature sensors. In besides to
control the actuation and deactivation of the actuators involved in the irrigation action,
consequently, the PCB shares the power lines with the power supply, it uses a standard
Fig. 5. Block diagram of the sensor node.
domestic power line (110V in México) to energize said actuators and control them
through the change of state of optocoupled relays. The use of batteries for the irrigation
node is added, to provide redundancy in power for the perception phase in case of
disconnection in the electric supply. The processing unit is carried out (as with the
sensor and master node) with the AtMega328p microcontroller and the wireless
transceiver as the nRF24L01+ (see the Fig. 6).
Fig. 6. Block diagram of the irrigation node.

To provide certainty in the irrigation task avoiding errors by duplication or desync

(in case of disconnection of the WSN or a voltage dip) a real-time clock block is
provided. This module is based on the DS3231 IC, which has an internal crystal
oscillator with temperature compensation. The DS3231 provides an accurate count of
time despite losing power since it has an extra battery that allows retains the time
counting, this effectively avoids failures or duplication in the irrigation in case the node
is not energized correctly.
Master Node. Finally, the master node is responsible for coordinating communication
between the nodes; this is achieved through the dynamic allocation of the nodes,
keeping the network active, and centralizing the data. When a frame is received in the
master node, the MCU process the frame and shapes it to the corresponding format
before and send it to the Server. In the same way, the master node obtains the con-
figuration information of the nodes inside the greenhouse through consumption of
REST services (see Fig. 7).
Fig. 7. Block diagram of the master node.
2.3 Server
The server application is developed on the single-board computer Raspberry Pi 3
model B+ development card with Raspbian operating system (based on Debian -
GNU/Linux) due to its processor, memory and radio transceiver characteristics based
on the IEEE 802.11 standard already included in the card. The design of the server is
divided into two parts:
Communication. Two forms of communication with the server were established to
guarantee the link, through the local area network using the Ethernet port, taking
advantage of the user’s infrastructure and through wireless communication provided by

the server when is configured as an access point (WAP) (see Fig. 8).
Fig. 8. Communication lines of the master nodes and the mobile application with the server.
Functionality. For the development of the web server, we use Django which is a high-
level application programming environment based on Python, Django allows us to
create web servers in a friendly way and create the REST services necessary for the
solution. The web application is constituted by Models, Uniform Resource Locators
(URLs), and Views. The information is stored and managed by the database
Fig. 9. Block diagram of server communication.

management system MariaDB, which offers the services of storage and control of the
tables. The relational model is selected due to the need to link the data of different
tables to establish the information of the crops, their belonging, and configuration (see
Fig. 9).
2.4 Graphical User Interface

For the development of the graphics interface, we decided to use the framework for
mobiles applications Xamarin.Forms through Microsoft Visual Studio IDE, which
provides a programming environment in C# language to make applications using native
resources, but with the option of sharing the code between the various mobile platforms
(Android, iOS and Windows Phone).
The Fig. 10 shows the state machine diagram of the User Interface that provides the
necessary views to add, configure, and eliminate the data of monitored crops, their
configurations, and user and application information.
Fig. 10. State machine of the views (activities) of the mobile application.
2.5 Payload Design

The payload format design was made according to each communication phase between
to each system element.
Intra-Greenhouse Communication. In the communication between the equipment
inside the greenhouse, the data is sent within the payload defined by the NRF24L01 +
transceiver, which emulates the BLE protocol frame (Fig. 11).
The information that reflects the values obtained from the environment are pre-
processed in the node prior to constructing the frame defining the data type as byte to
Fig. 11. Payload of the configuration message for irrigation node and sensor.
Table 1. Parameters included within the payload field of the nRF24L01+ frame.
Node type Parameter VMIN VMAX Precision Steps Data size
Sensor Temp 2 °C 35 °C 0.2 °C 166 1 byte
Hum 50% 85% 0.25% 141 1 byte
Ilum 0 60 Klx 10 lx 6001 2 bytes
Irrigation TempSol 10 °C 30 °C 0.1 °C 201 1 byte
pH 5 7.5 0.1 26 1 byte
make the memory space per parameter more efficient, the Table 1 shows data size of
the parameters monitored for hydroponic crops: Temperature (Temp), Humidity
(Hum), Light intensity (Ilum), temperature of the Nutrient solution (TempSol) and pH
level (Table 1).
For the communication of the message configuration by the master node to the
sensor and irrigation nodes, it is necessary to add to the payload (Fig. 11) the maximum
and minimum values of the parameters, their status (if enabled), the period in which the
sensing sample will be taken. Also, the data about start and end time in which the
configuration is valid, and in the case of the irrigation node, information referring to the
irrigation activity (irrigation periods and duration), so 21 bytes of data was defined for
data configuration message to be transmitted in this stage of design (see Fig. 12).
3 Development
This section briefly explains how the communication between the intra-greenhouse
network and the extra-greenhouse network is developed, as well as the structure of the
JSON (JavaScript Object Notation) messages that are used for the communication of
the master node with the server.
Fig. 12. Data structure of the JSONs sent from the master node to the server
3.1 Extra-Greenhouse Communication

The data coming from the sensor and irrigation nodes are extracted from the BLE frame
format by the master node and structured in a form for consumption of REST services
of the Server. About the Nodes configuration message, the master node requests the
data to the server and deserialize it to format in BLE frame structure to be sent to the
sensor and irrigation nodes.
3.2 User Interface – Server Communication

After performing the actions defined in the Server Views and the information is stored
in the database, it is required to request the data to the server to be showed in the user
interface on the mobile device. This data is sent in JSON format through the con-
sumption of the REST services. The Fig. 12 shows the data structure of the information
sent from the nodes to the server.
3.3 BLE Mesh Network

To implement the BLE mesh network, the nodes are configured in the following roles:
master node coordinates the routing configuration of the network allowing it to keep it
enabled, including define the identifiers of the irrigation nodes and sensors and
managing the retransmission of the frames among them. The Figs. 14 and 15 show the
flow diagram of the intra-greenhouse BLE mesh network established.
After initializing the nRF24L01+ transceiver and the sensors installed in the nodes,
the default range values are read from the internal MCU EEPROM; those values act as
extreme values for the monitoring of the environment (Temperature, Humidity, Illu-
mination, pH, Temperature of the nutrient solution). Likewise, the identification
information of the node (node type and serial number) and the configuration of the
control parameters of the system (sampling periods, times in which the configuration is
valid and node status) are read.
The configuration and management of the mesh network begin as the master node
dynamically assigns an identifier to the slave nodes (sensor node and irrigation node),
promoting the dynamic registration (commissioning) of them and the joining and
leaving of nodes to the network (see Fig. 13), thus allowing relocation of the nodes at
any time. The connection between nodes is updated periodically to maintain the link
between the nodes.
Finally, the operation of the slave nodes complies with the following steps: (see
Fig. 14).
Fig. 13. Flowchart of the initialization of the nodes for the creation of the BLE network.
Fig. 14. Flowchart of functionality of the Sensors, Irrigation and master nodes
1. Establish communication channels where the node receives configuration messages

from the master node or retransmit reading messages obtained from other nodes
(received messages cases S (Sensor) and R (Irrigation)).
2. If the node receives a message from the master node (message case C (Coordina-
tor)), the values in the payload are stored in the EEPROM of the microcontroller
dedicated to configuration values defined by the user.
3. Then the nodes perform the sampling of the environmental values through the
sensors according to the configured sampling period.
4. The MCU process the values obtained according to the previously defined ranges.
5. Create the frame with the payload in the format designed for the message destined
to the master node.
6. Establish the communication channel to the nearest node and transmit the message.
7. Go back to step 1.
4 Implementation and Results
The development and implementation of the system were carried out between the
facilities of the ESIME Zacatenco and a micro greenhouse in the town of Santa Isabel
Tola, the network of nodes was developed with the proposed sensors, a master node, a
server, and a prototype of the mobile application were developed (Fig. 15).
Fig. 15. Basic electronic diagram of Sensor Node and Irrigation Node
The configuration of the NRF24L01+ transceiver and behavior of the nodes to

deploy the BLE mesh network within the greenhouse was defined, in the Fig. 16 shows
the functionality of the nodes, in which the acquisition of the physical environmental
data is shown.
Fig. 16. Information capture of the sensor and irrigation nodes
The user utilizes the mobile application to enter the system to monitoring and
configure the state of crop actuators in a friendly way (Fig. 17).
Fig. 17. Startup and login pages of the mobile application.
Then select a crop to visualize, and the application displays the information of the
crop, its location, its assigned greenhouse, and the last recorded reading of the sensors
in the location (see Fig. 18).
Fig. 18. List of registered crops and display of the details of a crop
5 Conclusions and Next Steps
This system is developed focusing on low-cost technologies that allow its implemen-
tation by independent farmers, self-consumption and micro and small businesses,
which in most cases cannot acquire commercial systems due to its high cost. However,
we have observed that the flexibility and modularity of the system allows it to be
installed in many more places, for example, in urban and roof gardens or in indoor
hydroponic systems, which can have a significant impact on the reduction of land used
for traditional culture.
The above allows us to conclude that with minimal modifications this architecture
could be implemented in massive hydroponic farms, improving the monitoring systems
of these and with it the quality of the production. Finally, these types of modular IoT
architectures can revolutionize the industry, lowering the cost of infrastructure and,
soon, increasing the quality of life of the users.
The first stage of design was completed for the Intelligent system management for
hydroponic crops, the intends of the solution pretend to be an easy tool for farmers and
urban agricultures to improve their crops through the IoT technologies without high
investments.
Currently, we are working on the implementation in a real hydroponic greenhouse
to verify the correct functioning of the system. Based on the results of this imple-
mentation, the aim is to develop a machine-learning based algorithm that allows
adjusting the monitoring ranges automatically, to optimize crop quality and yield.
It is also proposed to implement an LTE-M modem that allows monitoring
remotely, as well as the sending of alerts by SMS, as an additional way of informing
the user of possible ongoing problems.
Finally, we are working on a second version of the system focused on medium-
scale crops, based on WSN, with greater coverage and flexibility, including a backup
server in the cloud, so that this new system can be managed anywhere with an Internet
connection.
References
1. Servicio de Información Agroalimentaria y Pesquera (SIAP). Uso de tecnología y servicios
en el campo Cuadros tabulares (2016)
2. World Bank, CIAT, CATIE. Climate-smart agriculture in Mexico. CSA country profiles for
Latin America Series. World Bank Gr (2014)
3. Oasis, S.: Manual de hidroponia, p. 32 (2002)
4. Patokar, A., Gohokar, V.: Precision agriculture system design using wireless sensor network,
pp. 169–177 (2017). https://doi.org/10.1007/978-981-10-5508-9_16
5. Verdouw, C.: Internet of Things in agriculture. CAB Rev. Perspect. Agric. Vet. Sci. Nutr.
Nat. Resour. 11 (2016). https://doi.org/10.1079/pavsnnr201611035
6. Azcón-Bieto, J., Talón, M.: Fundamentos de fisiología vegetal (2003)
7. Araceli, M.: Manual de hidroponia, Primera Ed. Instituto de Biología, Distrito Federal
(2014)
8. Kiani, F., Seyyedabbasi, A.: Wireless sensor network and internet of things in precision
agriculture. Int. J. Adv. Comput. Sci. Appl. 9, 99–103 (2018). https://doi.org/10.14569/
ijacsa.2018.090614
9. Minerva, R., Biru, A., Rotondi, D.: Towards a definition of the Internet of Things (IoT).
IEEE Internet Things 86 (2015). https://doi.org/10.5120/19787-1571
10. Lin, J., Yu, W., Zhang, N., Yang, X., Zhang, H., Zhao, W.: A survey on Internet of Things:
architecture. Enabling Technol. Secur. Priv. Appl. 4, 1125–1142 (2017)
11. Chiang, M., Zhang, T.: Fog and IoT: an overview of research opportunities. IEEE Internet
Things J. 3, 854–864 (2016). https://doi.org/10.1109/JIOT.2016.2584538
12. Yang, Y., Wu, L., Yin, G., Li, L., Zhao, H.: A survey on security and privacy issues in
Internet-of-Things. IEEE Internet of Things J. 4, 1250–1258 (2017)
13. Shi, W., Cao, J., Member, S., Zhang, Q., Member, S.: Edge computing: vision and
challenges. IEEE Internet Things J. 3, 637–646 (2016). https://doi.org/10.1109/JIOT.2016.
2579198
14. Razzaque, M.A., Milojevic-Jevric, M., Palade, A., Cla, S.: Middleware for internet of things:
a survey. IEEE Internet Things J. 3, 70–95 (2016). https://doi.org/10.1109/JIOT.2015.
2498900
Informatics Security
Learning and Automatization of Searching
in Digital Media, Based on Concurrency,
Weighing and Decay
David Cordero Vidal(&) and Cristian Barria Huidobro(&)
Vicerrectoría de Investigación, Centro de investigación en ciberseguridad CICS,

Universidad Mayor, Santiago, Chile
{david.cordero,cristian.barria}@mayor.cl
Abstract. The search for information of real interest for a particular topic has
always been a challenge, especially when the source of this information is on the
Internet, where there are many digital media, platforms, databases and from
which you can explore data. Large companies such as Google or Yahoo help us
find these sources of information through their search algorithms, but they are
not very effective when filtering a particular information, so it is necessary to
formulate a model that responds to this need, in the present The document
proposes a model based on 3 components, using information filtering tools,
word valorization techniques based on their concurrences, weights, anti-words
and a proposal to apply a word decay log within the information, in order to
maintain a “score” and avoid false weights (triggered weights) of the
information.
Keywords: Search Python Feed System Automatization Notification

Filtered out
1 Introduction
Information seeking through the network is a challenge nowadays. Internet has

transformed in one of the main basic information sources for many systems, platforms
and news media. Although there are tools which make easier to have access to
information sources (Google, Yahoo, etc), these do not provide a precise solution when
trying to find information of interest on specific topics. Some studies have researched
on the way of filtering significant information employing neural networks, machine
learning [1], IA [2], which solve, at a certain level, this issue.
The present investigation proposes a research method of digital information based
on three components (obtainment, classification and acceptance). The objective is to
filter interesting information for a person or organization, by determining the average
time a key word appears within a piece of information based on a given dictionary.
Later, to generate a proposal of key words which were not considered in order to define
a filter drawn from each key word’s weighted value within an information set.

https://doi.org/10.1007/978-3-030-33229-7_27
326 D. Cordero Vidal and C. Barria Huidobro
2 Proposed Method for Information Searching
The investigation is based on a proposal of information searching and filtering drawn

from three components: to know (1) information obtainment, (2) information classi-
fication, and (3) information acceptance. Each one needs tools and functionalities
which are also determined by some variables.
2.1 Obtainment Information Component

This first component determines the digital informative media from which the infor-
mation will be collected, by means of employing an algorithm drawn from the
Universal Feed Parser library [3]. This is a Python module that works to download and
analyze syndicated feeds; it supports RSS 0.90, Netscape RSS 0.91, Userland RSS
0.91, RSS 0.92, RSS 0.93, RSS 0.94, RSS 1.0, RSS 2.0, Atom 0.3, Atom 1.0, and CDF
feeds. This library will allows us to base this investigation on a single algorithm, as the
data delivery is structural and unchangeable, independently of the information source
(shown on Fig. 1).
Fig. 1. FeedParser functionality.
In this way, it is possible to determine the variables that define the obtainment
process. These are found on all digital media which fulfils the condition of delivering
information in the Feed RSS format (Really Simple Syndication) [4].
• Obtainment Variables: RSS Feed Structure: From this RSS structure taken from the
FeedParser library, we will base the information searching on collecting the data
now shown. These will change their content as the information source feeds update,
but the structure always remains unchangeable.
– feed.tittle
– feed.url
– feed.content
– feed.tags
• Information Digital Media: The Information Digital Media (news websites, blogs,
etc) will deliver the information requested by the model. These could be classified
by topics, and the collecting condition will be given by the information delivery
format (Feed RSS).
Learning and Automatization of Searching in Digital Media 327
• Feed Update Time: The different information digital media update their feeds when
modifying the information. For example, every time a digital newspaper publishes a
news, the feed changes and generates new information captured by the model.
• Collecting Cycle Time: This refers to the time the model will last to collect all the
digital information media before starting again.
• Total Collecting Time: The total time employed to collect the information
(Table 1).
Table 1. Variables of the information obtainment.

Variable Measure unit
Feed Text (it changes according to the information digital media) tittle, tags
and information body
Information digital Number of websites registered
media
Collecting cycle Minutes
time
Total collecting Days
time
Feeds update time Minutes
2.2 Classification Information Component

Once the information has been provided by the obtaining component, it is necessary to
classify and filter by key words; this determines two filters: the first is by tittle and tags,
the second is by the information body reading. From this, the following variables are
determined.
Classification Variables, words: These are found within the tittle, tags and infor-
mation body. They have a semantic meaning, and it is possible to determine their topic
of discussion through their study.
Stopword: The term Stopword [5] refers to words that lack meaning, also known as
empty words. Usually they refer to articles, pronouns, conjunctions, prepositions,
among others. Likewise, to eliminate these from titles, tags and information body
optimizes the algorithm route when looking for key words.
Key Words: These correspond to words related to the topic of interest. Google, one
of the leading sources of information searching on Internet, puts at people’s disposal
tools that help them on their searches. If we focus on these searching tools, the first that
comes to our minds is the browser [6], which its algorithm is secret (private source
code). Nonetheless, one of its revealed characteristics is the key word searching, which
is determined by the frequency or location these have within a website. Key words will
be defined by means of the topic of interest itself or by the person’s own judgement [7].
It is important not to define words that are considered to be “Stopwords”, as these
will not be considering when filtering the information. This is because these words are
considered meaningless by itself.
Concurrency: The concurrent value will be given by the time a key word shows on
the information (Table 2).
Table 2. Variables from the information-collecting component.

Words Number of words
Key words Number of key words
Stopword Number of Stopwords
Concurrency Number of times
Classification Tools: To classify the information is necessary to employ tools that

make easier the component’s process. In this investigation, the following will be
employed.
Dictionaries: Repository of related words. Each dictionary contains its own key
words [8]. The weighted value (the importance of a word within a dictionary) of each
word will be given by the topic of interest, as the same word can have another
relevance within other context.
For example, the word “boy” could appear in many dictionaries or categories. If
this is found on a dictionary of children’s games, its weighted value will be higher than
if it is found on one of table games. In this way, the model works finding information
by means of organized word groups that have a closeness of meaning with the topic of
interest.
Natural Language Toolkit (NLTK): The use of the library NLTK [9] that incor-
porates a support for thousands of lexical corpuses and trained modules, such as
wordnet. Also, it includes tools to process, derive, label and conduct semantic analyses,
which clean the text from Stopwords and vectorize only those that give meaning to the
information. This finally optimizes the vector’s final route over which the model is
based.
2.3 Acceptance Information Component

Once the information has been classified, the model has two acceptance criteria. The
first is the key words’ weighing in the information body, which refers to the average of
the weighted sum drawn from the filtered information plus the expert’s judgment. The
second is the concurrency of key words about the given information. For this, the
acceptance procedure is conditioned to the following variables.
Acceptance Criteria Based on Concurrency: To define the acceptance criteria is
necessary to define the quantity of key words that will be shown on the title, tags and
information body. From this, a dictionary will be prepared based on the topic of interest
and the words related to it. Under this condition, the information will be filtered, and
for the purpose of this study, two filtering conditions were defined. The first is based on
the concurrency of key words in the title and tags; the second on the information body
(Table 3).
Table 3. Variables of the weighing component

Acceptance criteria Number of key words
3 Component Metrics
The sample taking is conducted to register the metrics of each component, in which the
values of the identified variables are obtained. This is conducted by a determined
hardware and software, which are mentioned as following:
Hardware:
• Intel I5 8th Generation
• 8 GB Ram
• 256 GB SSD
The software. Libraries employed:
• Linux Kubuntu 16
• Python 3.6
• Framework Django 2.1
• Virtualenv 17.0
• PostgreSQL 11.2
• NLTK library, Text Blob
• FeedParser 5.2.0
3.1 Metrics of the Obtainment Information Component

The samples are obtained from a hundred information digital media (website URL,
which provides the Feed information). These media are categorized as informative
(national and international news). For example, if the topic is “Armed Forces”, the
dictionary will have the same topic. In this way, we find 120 key words that relate to
the Armed Forces (Table 4).
Table 4. Variables of the obtainment component

Variable Measure
Feed It changes depending on the collected fee (Title, Tags, and Body)
Information media 100 URLs
Collecting cycle time 11 min
Feed update time 15 times average
Each collected fee changes from the news media (the news title, the body and the
related tags). The update average time of these 100 media is of 15 min. This is variable
because it changes depending on the news’ contingency. At the moment of this study,
the contingency developed normally, without existing a high number of interesting
news in the news media. The model lasts approximately 11 min to track all the reg-
istered websites. From these metrics, 17.000 feeds were collected and these will sustain
the information classification component.
3.2 Metrics of the Classification Information Component

The obtained metrics come from the number of key words in the information titles and
tags. In the following table, it is shown that the more key words are found in the first
filter, less news are registered. However, it is also shown the loss of many candidates
news (example: news that can be of interest for the topic proposed). From now on, the
filter of 01 key words was chosen to be worked with as first filter (Table 5).
Table 5. Filter of key words over 17.000 news from a 100 information media
Number of key words Number of candidate news Percentage of total news
01 4.320 25,41%
02 338 1,98%
03 44 0,25%
3.3 Metrics of the Acceptance Information Component

The present investigation based its information acceptance logic on the representation
procedure of Bag-of-Words texts [10], using the NLTK library. In this way, the
information corpus selected is narrowed and represented as a vector of constant length,
which reflects the onset of each term in the text.
Two forms are proposed for accepting a candidate piece of information: (1) under
the criteria of the sum of total concurrency of key words of the dictionary in the
information body, and (2) the total weighted value of the information body. The last is
only a proposal, and object of future study that is further explained in the further studies
section.
Acceptance Criteria based on Concurrency: As it is shown in the following table,
the more number of key words are found in an information body, the percentage of
successes increases (example: interesting news about the topic), and as counterbalance,
less news are declared to be candidates. The loss of news is relevant to understand the
model leaves out information that could be of interest as from the acceptance criteria of
04, 05, 06 and 07 key words in the information body (from a basis of 120 key words in
the dictionary). As conclusion, it could be said that from 6 key words on, the degree of
success is higher than the filter loss (Table 6).
Table 6. Information acceptance over 4.320 news

Number of Key Number News Number of filtered Percentage of
Words in the of filtered of news from the total Interest about
information corpus news Interest information filtered news
04 1344 413 7,90% 30,72%
05 763 411 17,66% 56,22%
06 466 405 10,7% 86,10%
07 287 264 6,64% 91,98%
4 Model Proposal
From the components described previously, we propose the following model based on
three components (Fig. 2), and a flow chart which explains the model’s addressing
(Fig. 3).
Fig. 2. Proposed model based on three components
Fig. 3. Processes of the proposed model

5 Continuous Learning
Once the system has been sustained with the information of interest, it is possible to
automatically determine new key words, example: words not considered by the
expert’s judgement at the beginning of the procedure. With the library NLTK (Natural
Language Toolkit), all corpuses available can be vectorized, stopwords erased and
significant words are selected. Likewise, it can be determined which words are more
concurrent so to propose them to the dictionary indexation.
Consideration: It is the value or importance a key word has within a dictionary, and it
relates to the topic of interest. In this way, the words are prioritized according to their
importance in the topic. The same logic is employed by the procedure based on
representation of Bag-of-Words’ text [10], and used by the NLTK library to assign the
relevance meaningless or too repetitive words could have (Fig. 4).
Fig. 4. Key words proposal
Even the acceptance criteria could be given by a collection of news of interest,

which allows to get the individual values of each information and calculate an average
result that defines a base line from the samples available. Mathematically speaking, this
logic is represented for the following equation:
5.1 Acceptance Criteria Based on Weighing

By the use of these weightings, we can reduce the false positive in the collection of
feed. This procedure applied as a second filter to the acceptance criteria will be given
by the sum of words’ weightings from the dictionary applied to the corpus or infor-
mation body (RSS content). As an example, the following table shows the basis of a
dictionary with the words’ values and weightings (Table 7).
Table 7. Value and weighting of key words

Key word Value Weighting
Boy High 0.4
Park Medium 0.2
Bicycle Medium 0.2
Boat Low 0.1
Ball Low 0.1
The total weighing dictionary has to be equal to 100% in the amount of words
registered to define the correct prioritization. The value of each word depends on topic,
as the same word in different topics could have different weighing values (importance).
The weighing value could be assigned by the expert’s judgement.
Decay Logarithm: If there are repeated words within the body information, a
logarithm of decay will be applied [11]. The value for this logarithm will be obtained
from the number of manifestations shown in the vector. This procedure is conducted
because a single key word can alter the acceptance criteria, as many concurrencies can
show a high weighing followed by a value decay within the total sum of words found.
If the vector of the frequency (repetitions) of the dictionary words f^(t) in a given
text t, then to calculate the score of the text it would be enough to calculate the next
sum:
sum fword i of the dictionaryg d iðf i^ ðtÞÞ
where f_i^(t) is the number of times the word number i in the dictionary appears in the
text t and d_i are functions of the form:
d iðf i^ ðt)) = 0 when f i^ ðt) = 0 ð1Þ
d iðf i^ ðt)) ¼ a i þ b i log(f i^ ðt)) when f i^ ðt) 1 ð2Þ
where a_i and b_i are parameters that can be calibrated.

Anti-Key Words: A dictionary of anti-key words is also defined; that is to say, words
that were not considered within the information context. For example: if the topic of
search is Politics, the word Football, considering the information corpus, will reduce
total sum of key words and their total weighing.
Nowadays, it is fundamental to obtain information from different digital media. The

necessity of maintaining oneself informed determines the opportunities in the digital
area or 4th generation [14]. For this reason, blogs, social networks, online newspapers
become the main information sources and are the ones which valorize the same.
We think it is crucial to develop a model that allows getting information of real
interest in no time, but this is only possible if we have a data source that has been
filtered correctly.
We believe that some information could be crucial for people or organizations in
particular, and certainly, it is possible to narrow these search processes when pre-
existing tools are employed correctly.
From the collection of massive information pieces of real interest on a defined
topic, with the support of NLTK and Text Bob, it is possible to obtain useful infor-
mation, determine sub-topics or tendencies not previously considered, in order to relate
this information to future events. Keeping the logic of the proposed model, we could
even conduct this search towards social networks.
An example will be Twitter [12], which provides an API with which we can look
for Tweets related to a topic. Another will be the tools of Google, particularly Google
alerts, which provide information available on the website through sides that have not
been registered in the system (digital information media).
The integration of a system that connects to any API and sends real-time notifi-
cations at the moment of finding a piece of information is crucial. Libraries such as
Telebot (Telegram) allows to manage these notifications directly from the source code.
It is necessary to create only a BOT from Telegram [13], to include the application
token key directly to the source code in order to send the messages with the results.
References
1. Dua, S., Du, X.: Data Mining and Machine Learning in Cybersecurity. Auerbach
Publications, New York (2016). 9780429063756
2. Fawaz, S., AbuZeina, D.: Toward an enhanced Arabic text classification using cosine
similarity and Latent Semantic Indexing. J. King Saud Univ. – Comput. Inf. Sci. 29(2), 189–
195 (2017)
3. Feedparser. https://pythonhosted.org/feedparser/. Accessed 06 May 2019
4. Bharathi, S., Geetha, A., Sathiynarayanan, R.: Sentiment analysis of twitter and RSS news
feeds and its impact on stock market prediction. Int. J. Intell. Eng. Syst. (2017). https://doi.
org/10.22266/ijies2017.1231.08
5. Saif, H., Fernández, M., Yulan, H., Harith, A.: On stopwords, filtering and data sparsity for
sentiment analysis of Twitter. In: Proceedings of the Ninth International Conference on
Language Resources and Evaluation, LREC 2014, pp. 810–817 (2014)
6. Prat, M.: Posicionamiento web: estrategias de SEO: Google y otros buscadores. s.l.: Eni
(2016). 2409004903
7. Cabero, J., Llorente, M.: The expert’s judgment application as a technic evaluate information
and communication technology (ict), Revista de Tecnología de Información y Comunicación
en Educación, vol. 7, no. 2 (2013)
8. Díaz, I., Suárez, S., Sidorov, G.: Creation and evaluation of a dictionary tagged with
emotions and weighted for Spanish, Onomázein: Revista de lingüística, filología y
traducción de la Pontificia Universidad Católica de Chile, no. 29, pp. 31–46 (2014). ISSN-e
0717-1285
9. Perkins, J.: Python 3 Text Processing with Nltk 3 Cookbook (2014). ISBN-10: 1782167854
10. Hasan, K.S., Ng, V.: Stance classification of ideological debates: data, models, features, and
constraints. In: Proceedings of the Sixth International Joint Conference on Natural Language
Processing, pp. 1693–1701 (2015)
11. Kumar, S., Morstatter, F., Liu, H.: Twitter Data Analytics. Springer, Heidelberg (2013).
https://doi.org/10.1007/978-1-4614-9372-3. ISBN 1461493722
12. Setiaji, H., Paputungan, I.: Design of telegram bots for campus information sharing. In: IOP
Conference Series: Materials Science and Engineering, vol. 325, p. 012005 (2018). https://
doi.org/10.1088/1757-899x/325/1/012005
13. Salas Alvarez, D.: Revolución 4.0, RII, vol. 4, no. 2 (2016)
Security Hierarchy for the NDN Network:
Description and Experimentation
P. David Arjona-Villicaña(B) and Alejandra G. Silva-Trujillo
Facultad de Ingenierı́a, UASLP, San Luis Potosı́, Mexico

{david.arjona,asilva}@uaslp.mx
Abstract. Information centric-networks propose a new architecture for

the Internet where IP addresses are replaced by names that represent
the information or service that users request. The Named Data Network-
ing (NDN) project is seeking to implement and experiment one type of
these networks. A fundamental principle of NDN is that all the infor-
mation needs to travel through the network using a secure hierarchical
infrastructure. This article describes this security hierarchy and, using
experiments, shows its implementation in two test networks.
Keywords: Information centric networks · Distributed systems ·

Network security
1 Introduction
The Internet is a worldwide data network whose main objective is to allow pairs
of computers to exchange information. In order to identify each computer, it
becomes necessary to employ a unique number, the IP address. However, most
users do not need to know their own devices’ address, nor the procedures required
to exchange information in this network. These procedures are relegated to dif-
ferent protocols, like TCP and UDP. Since what it is important to the Internet’s
users is to obtain information and services, there are currently proposals to
implement new network architectures that are oriented to obtain, transmit and
deliver information, instead of establishing a connection between two computers.
This shift of the main architectural objective of the Internet has been named
Information-Centric Networking or ICN [1,2].
The main objective of ICNs is to implement a network infrastructure that
is better suited on how users currently employ the Internet, which are more
interested in accessing information, rather than its physical location. Special
attention is given to content distribution, security and mobility support. There
have been different proposals to implement ICNs. Examples of these are Data-
Oriented Network Architecture (DONA) [3], Publish-Subscribe Internet Tech-
nology (PURSUIT) [4], Scalable and Adaptive Internet Solutions (SAIL) [5] and
Named Data Networking (NDN) [6].
Named Data Networking or NDN is a type of ICN that is currently being
developed and implemented by the NDN Project. This project is a team effort
https://doi.org/10.1007/978-3-030-33229-7_28
336 P. D. Arjona-Villicaña and A. G. Silva-Trujillo
Fig. 1. The main functional elements of the current Internet and NDN.
by many universities and research laboratories around the world, including the
Universidad Autónoma de San Luis Potosı́ (UASLP) [7]. Currently, participants
to this project are trying to implement the main functionalities of an NDN node,
for example: hierarchical name structure, security methods and procedures, rout-
ing, forwarding and in-network storage [6]. Others are working on applications
that need to employ some of the functionalities described before. The main idea
behind these developments is to prove NDN functionality by testing it in a setting
that is as close to a real network as possible. Therefore, the NDN project is also
implementing a global network to test the protocols, services and applications
under development [8]. This network includes a node at UASLP.
Another of the main objectives of NDN is to maintain a structure that is as
simple as the one currently implemented for the Internet. This has been called the
thin waist model (Fig. 1). For today’s Internet, the simplest (thinnest) functional
element is the IP packet; for the case of NDN, is the Content or the name of the
information, which is sent through two different types of packets, Interest and
Data.
Addresses in an NDN network are the name of the information or service
that the user requests [6]. These addresses follow a hierarchy which enables
names that are more specific to have priority over more generic ones. For exam-
ple, a name like /uaslp/ingenieria/computacion has precedence over /uaslp or
/uaslp/ingenieria. This is similar to how current routing protocols work, where
a network using a /24 mask (a set of specific IP addresses) has priority over
another using a /20 mask (a less specific set of addresses). NDN employs only
two types of packets: Interest and Data. Interest packets are used to request
information and Data packets are used to answer each Interest packet. Neither
of these packets is associated with the computer that originally requested the
information or the one that generated the Data packet.
Security Hierarchy for the NDN Network 337
In an NDN network, routers maintain a database which temporarily stores

the received Interest and Data packets. Each Data packet must satisfy an Interest
packet that has been previously received. Since no packet is associated to the
computer that originally generated it, it is possible for a router to receive requests
for the same Interest from different devices or network interfaces, which the
router satisfies by sending the same Data packet that has been temporarily
stored in its database. Following this procedure, a router may send the same
information which is requested by different users in a short period of time. This
will reduce the traffic in the network since the router only needs to receive one
Data packet to satisfy many Interest packets. This also means that NDN is not
connection-oriented and that the network has a built-in memory that allows it
to distribute information that is constantly or simultaneously being requested.
Since NDN introduces a new paradigm for the architecture of the Internet,
it becomes necessary to define the other elements that form this network, for
example, security. The Internet was originally based on trust and security mech-
anisms were added as needed [9]. NDN follows the opposite approach: it is a
network that requires security mechanisms to prevent that Interest and Data
packets are modified as they travel through the network [10].
Because security is a requisite to exchange messages between nodes in an
NDN network, it becomes necessary to understand and being able to set up
security before other applications and protocols can be used or tested. Therefore,
the main contribution of this paper is to provide two experiments that show how
to configure security in an NDN network. These experiments are described at
Sect. 2.
Other elements that need to be defined for NDN are routing protocols. Cur-
rently, a protocol based on Dijkstra’s algorithm [11], the Named Data Link State
Routing Protocol (NLSR) [12], is being used. However, before this or any other
protocol exchanges routing information in an NDN network, it is necessary to
define and configure a security infrastructure. This article describes how to con-
figure this infrastructure (Sect. 2.1) and includes a pair of experiments that show
how to perform these configurations (Sects. 2.3 and 2.4). After describing these
experiments, the article provides conclusions at Sect. 3.
2 NDN Security
This section first describes the NDN security hierarchy (Subsect. 2.1). Followed
by a description of the software needed for implementing the experiments (Sub-
sect. 2.2). Finally, two experiments are described in Subsects. 2.3 and 2.4.
2.1 NDN Security Hierarchy
NDN follows a hierarchical security structure (see Fig. 2). A root is at the top of
this hierarchy and is the entity in charge of certifying each site in the network.
Then, each site certifies one or more operators which, in turn, certify each router
in the network. Each one of these different entities must be defined and certified
Fig. 2. Example of NDN’s hierarchical security structure.
in an NDN network. For networks that have few nodes, it is possible that one
router simultaneously assumes the function of more than one of these entities.
It is important to consider that this security hierarchy needs to be defined and
configured before any other protocol or application can exchange messages. This
includes even routing protocols, like NLSR.
Figure 2 shows an example of a security structure for and NDN network with
two sites. Site 1 employs two operators: operator 1 is responsible for router 1
and router 2, while operator 2 is responsible for router 3 and router 4. Site 2 has
a simpler configuration since it only has one operator (operator 3) which only
handles router 5. This hierarchy is independent of how the routers are connected
between themselves and, as it was said before, more than one entity could be
defined in a single router or node.
In order to implement this hierarchy, it is necessary to exchange keys and
certificates using the following procedure: The dependent entity (the one that
is below in the hierarchy) generates a key which is then copied to the parent
entity (the one that is above). The parent entity employs the key as an input to
generate a certificate, which is then copied back to the dependent entity. Finally,
the dependent entity installs the certificate. This procedure is shown in Fig. 3.
The only exception to this procedure is the root, which is the only entity allowed
to self-certify its key.
2.2 Previous Work to the Experiment
Before performing the experiments described in Sects. 2.3 and 2.4, it is necessary
to install ndn-cxx and NFD in each node in the test network. The ndn-cxx library
is a set of programs developed in C++ that implement different NDN support
applications. The NDN website [13] includes information and instructions on
how to install this application’s library. NFD is an application that implements
an NDN forwarder node that supports the NDN protocol and is able to receive
and send Interest and Data packets. The NDN site [14] includes instructions on
Fig. 3. Generating and installing keys and certificates in NDN.
how to install this application and a description of NFD’s architecture may be

found at its Developer’s Guide [15].
To verify that the security hierarchy has been successfully configured, it is
necessary to execute an application that employs this hierarchy. The NLSR
protocol is an example of these applications. But before installing NLSR, the
Chronosync and Psync applications must also be installed. More information
about installing NLSR, Chronosync and Psync may be found at [16].
It is also recommended to review the beginners’ installation guide [17] which
includes all the necessary steps to install ndn-cxx, NFD and NLSR in a computer
running the Linux Fedora operating system.
2.3 Experiment 1
This experiment shows how to configure the security hierarchy for the sim-
plest NDN network: two routers communicating between themselves (see Fig. 4).
This network has two routers, R1 and R2, which connect to each other using a
crossover Ethernet cable. R1 is configured to be the root, site, operator 1 and
R1, while the other computer is only configured to be R2. The commands needed
to configure R1 are included below.
1. The following commands configure R1 as the root:

ndnsec-key-gen /ndn/ > root.key
ndnsec-cert-dump -i /ndn/ > root.cert
ndnsec-cert-install -f root.cert
Fig. 4. Topology and configuration for first experiment.
2. The following commands configure R1 as the site:
ndnsec-key-gen /ndn/edu/uaslp > site.key

ndnsec-cert-gen -s /ndn/ site.key > site.cert
ndnsec-cert-install -f site.cert
3. The following commands configure R1 as the operator 1:
ndnsec-key-gen /ndn/edu/uaslp/%C1.Operator/op1 > op1.key

ndnsec-cert-gen -s /ndn/edu/uaslp op1.key > op1.cert
ndnsec-cert-install -f op1.cert
4. The following commands configure R1 as a router:
ndnsec-key-gen /ndn/edu/uaslp/%C1.Router/R1 > R1.key

ndnsec-cert-gen -s /ndn/edu/uaslp/%C1.Operator/op1 R1.key
> R1.cert
ndnsec-cert-install -f R1.cert
Since R1 is already configured as the root, site and operator 1, R2 only needs
to be configured as a router whose parent entity is the same operator (operator
1). The commands included below need to be executed in R2 and R1 to perform
this configuration.
1. R2 generates its own key:

2. The key is copied to operator 1 (R1) and it is then certified using the following
command:

> R2.cert
3. The certificate is copied back to R2 and then it is installed using the following
command:
Fig. 5. Extract of NLSR log showing successful key registration and message exchange.
After finishing the previous procedure, it is necessary to execute NLSR and

verify that the configuration has been successful. Setting up and running this
routing protocol is beyond the topics covered by this article, hence, readers
may refer to the NLSR web page [16]. However, Fig. 5 shows an extract of a
log generated by this application, where it is possible to read that the different
keys have been successfully registered and that messages have been exchanged
using NLSR’s Hello protocol. Since the logs at Fig. 5 do not show error messages
related to key registration and is able to exchange packets, the security hierarchy
has been set up correctly.
2.4 Experiment 2
To verify the robustness and generality of Experiment 1, a second experiment

with a more complex structure and network topology was performed. The topol-
ogy and structure for this experiment is shown in Fig. 6. This topology includes
five routers which will perform the following roles: R1 is configured to be the
root, site, operator 1 and R1; R2 is configured to be operator 2 and router R2;
R3 is configured as a router whose parent is operator 2; R4 and R5 are configured
to be routers whose parent is operator 1.
Fig. 6. Topology and configuration for second experiment.
To configure the security hierarchy for R1, it is only necessary to repeat the
same steps that were shown at Sect. 2.3 for R1.
The commands included below may be used to configure security for R2.
1. Operator 2 generates its own key:
ndnsec-key-gen /ndn/edu/uaslp/%C1.Operator/op2 > op2.key
2. The key is copied to the site (R1) and it is then certified using the following
command:
ndnsec-cert-gen -s /ndn/edu/uaslp op2.key > op2.cert
3. The certificate is copied back to R2 and it is then installed using the following
command:
ndnsec-cert-install -f op2.cert
4. The following commands configure R2 as a router:
> R2.cert
To configure R3, R4 and R5 it is necessary to repeat the same steps that
were used at Sect. 2.3 for R2. However, these instructions need to be adjusted
to consider that R3’s parent is operator 1, while the parent for R4 and R5 is
operator 2.
After performing the security configuration for the five nodes in this network,
NLSR was executed again and it was verified that all the computers exchanged
messages and were able to generate a routing table.
3 Conclusions
ICNs are a proposal to redefine the Internet from a network that allows to estab-
lish communication between two computers, to an infrastructure whose main
objective is to obtain information from the network. This, in turn provides the
opportunity to examine new ideas and to reflect on the fundamental principles
that allowed to construct today’s network.
For the specific case of NDN networks, security mechanisms are required to
guarantee that Interest and Data packets are not modified as they travel through
the network. This paper has shown how to properly configure this security using
two different test networks. This correct configuration allows the implementation
of an Internet which should be safer by definition.
Besides the topic of security, there is a large number of protocols and appli-
cations under development and experimentation for NDN. These developments
represent an opportunity to contribute to a project which may have a signifi-
cant impact in the design of new architectures for the Internet. However, engi-
neers interested in contributing to the NDN project, first need to understand
its security hierarchy. The experiments included in this article are a first step to
understand this hierarchy and its main concepts.
The authors of this article are trying to propose new routing strategies for
NDN and this article shows the path that may be followed in the next stages of
research.
Acknowledgments. The authors would like to thank Lan Wang and Ashlesh
Gawande form the University of Memphis which helped to understand and clarify
many of the concepts exposed in this article.
References
1. Ahlgren, B., Dannewitzh, C., Imbrenda, C., Kutscher, D., Ohlman, B.: A survey
of information-centric networking. IEEE Commun. Mag. 50(7), 26–36 (2012)
2. Xylomenos, G., et al.: A survey of information-centric networking research. IEEE
Commun. Surv. Tutor. 16(2) (2014)
3. Koponen, T., et al.: A data-oriented (and beyond) network architecture. In: ACM
SIGCOMM, pp. 181–192 (2007)
4. FP7 PURSUIT project. http://www.fp7-pursuit.eu/
5. FP7 SAIL project. http://www.sail-project.eu/
6. Zhang, L., et al.: Named data networking. ACM SIGCOMM Comput. Commun.
Rev. 44(3), 66–73 (2014)
7. NDN Testbed Status. http://ndndemo.arl.wustl.edu/
8. NDN Traffic Map. http://ndnmap.arl.wustl.edu/
9. Bellovin, S., Schiller, J., Kaufman, C.: Security Mechanisms for the Internet, RFC
3631 (2003)
10. Zhang, Z., et al.: An overview of security support in named data networking, NDN,
Technical report NDN-0057 (2018)
11. Dijkstra, E.: A note on two problems in connection with graphs. Numerische Math-
ematik 1(1), 269–271 (1959)
12. Lehman, V., Mahmudul Hoque, A., Yu, Y., Wang, L., Zhang, B., Zhang, L.: A
secure link state routing protocol for NDN, NDN, Technical report NDN-0037
(2016)
13. ndn-cxx: NDN C++ library with eXperimental eXtensions. http://named-data.
net/doc/ndn-cxx/current/
14. NFD: NDN Forwarding Daemon. http://named-data.net/doc/NFD/current/
15. NDN Project Team, NFD Developer’s Guide, Technical report NDN-0021 (2016)
16. NLSR, Named Data Link State Routing. http://named-data.net/doc/NLSR/
current/
17. Arjona-Villicaña, P., Renterı́a-Vidales, O., Gawande, A.: A beginner’s guide
to installing and testing NLSR in Fedora. http://named-data.net/doc/NLSR/
current/beginners-guide.html
Methodology for Malware Scripting Analysis
in Controlled Environments Based on Open
Source Tools
Diego Muñoz(&), David Cordero(&),

and Cristian Barría Huidobro(&)
Centro de Investigación en Ciberseguridad, Universidad Mayor, Santiago, Chile

diego.munoze@mayor.clm,
{david.cordero,cristian.barria}@mayor.cl
Abstract. In today’s interconnected world, there is a latent threat called mal-

ware or malicious software. Different variations of these polymorphic and
metamorphic malware continue to evolve, even becoming large industries called
Malware as a Service (MaaS) [1]. This combined with the large number of new
technologies has evolved along with different threats, which can seriously
damage from a workstation, to large network architectures [2]. In order to face
it, it is necessary to be able to analyze and understand its operation, for this
reason to carry out this task a defined methodology is necessary. This paper
proposes a methodological structure for working with malware scripting, for
which a detailed example of practical application in a controlled environment is
illustrated. After the analysis of the results obtained, a concept map is offered
with the stages and activities related to the proposed methodology.
The present investigation provides an adequate look for the rapid analysis of
malicious scripts, which allows decisions to be made during situations of IT
crisis, which in turn will be the basis for a thorough further analysis.
In order to start with any type of analysis, it is important to establish a
working methodology or framework to be able to carry out a sample study of
some type of malware scripting, also considering identifying its classification,
based on “Malware Analysis and Classification: A Survey” [3].
Keywords: Malware scripting Static analysis Dynamic analysis

Methodology
1 Introduction
To take advantage of different security flaws or breaches of some system, cyber-

criminals focus on the end user, as Jorge Mieres points out in his whitepaper about
computer attacks, demonstrating how cybercriminals take advantage of exploiting any
weaknesses or failures in the systems [4] also considering use social engineering, as
Kevin Mitnick details in his book “The Art of Invisibility” [5], applying different
techniques so that victims can provide the necessary data for a future attack, which is a

https://doi.org/10.1007/978-3-030-33229-7_29
346 D. Muñoz et al.
cause of concern for cybersecurity managers who must generate computer security
awareness instances or culture.
Many of the attacks that occur today are based on malicious software and the
different variations that may arise from a single threat, it is for this reason that the need
for malware analysis is born, which in definition could be considered as the art of
dissecting malicious software, to understand how it works, how to identify it and thus
be able to defeat it.
Malicious codes or scripts will continue to be one of the main threats to the modern
world, becoming an endless battle between those who write new scripting malware and
those who seek to detect their execution in time and avoid all types of fraud through
cyberspace. The exploitation of systems through the infection of scripting malware also
propagated by phishing, are the perfect mix to continue deceiving the end user and to
penetrate even the most sensitive information we store.
2 Proposed Method Analysis
Currently there are different methods to perform different activities framed within the
field of cybersecurity and cyber defense, one of these disciplines is malware analysis.
As a way to help researchers and analysts on this matter, the present analysis of
malicious software is focused on malware scripting, that is very simple code sequences
that automate malicious tasks in real time and that do not need to be compiled as is the
case with files executables (.EXE). These can infect, download, redirect or slow down a
computer, in addition to not being discovered developers obfuscate the code to make
them undetectable and thus not be discovered by antivirus engines.
Some languages to perform this type of malware can be used different program-
ming languages, such as Visual Basic Script (VBS), Java Script (JS) or Python (PY).To
carry out this type of methodology, the following 5 stages are proposed:
2.1 Stage 1: Sample Identification

Every time we face a sample it is necessary to know what kind of file we are analyzing,
some try to deceive the victim by posing as images, folders or any other extension that
a normal user would consider harmless.
Upon receiving a sample of the malicious software we can rely on utilities such as
the “file” command that brings most of the GNU/Linux operating systems, which is
responsible for testing the file we want to evaluate and returns the information of both
its type and its format.
Already knowing the file type it is necessary to obtain its metadata, identifying the
size in bytes, date of creation, date of modification, among other characteristics, as can
be evidenced in the following (Table 1):
Methodology for Malware Scripting Analysis in Controlled Environments 347
Table 1. Data that must be identified

Data Example Note
Type Executable Can also be scripts
Format .EXE .AVI, .MP4, .PDF, .JPG, etc.
Weight 285.13 KB Can also be defined as MB, GB, etc.
Hash 8d357c0f0cc37dd573a2ba57b8 Can also be used for online analysis
Date 02-jun-2019 Creation and/or modification date
2.2 Stage 2: Static Analysis

Having the sample relevant background, we proceed to perform a static analysis of the
malicious code. The static analysis allows us to examine the sample code in a safe
environment, without the need to execute it, so we can generate hypotheses of what its
effects are or will be once executed.
In order to carry out this activity, reverse engineering is performed, that is, the
process to determine the operation by disassembling each component if necessary. For
this reason it is important that the analyst understands both high and low level lan-
guage, in order to carry out an analysis that allows conclusions to be drawn while
advancing in the study of malicious code, especially on low level language since it is
precisely this with which reverse engineering is performed, understanding in the same
way that both levels of language are interpreted by the CPU in a machine language. For
this reason it is essential for an analyst to know different types of programming lan-
guages [7, 8], which helps to better understand how the developer built the scripting
malware (Fig. 1).
Fig. 1. Levels of abstraction to interpret.
This type of analysis happens when it is necessary to calculate the different hashes
that the sample can have, such as its MD5 hash, SHA1, among others, it is also possible
to rely on integrated development environments (IDE), or it can even be studied
quickly with any text editor, since they do not need to be compiled for execution.
After the background or data obtained, it is important to look for information in open
sources, in order to know if this threat has already been studied by other researchers.
2.3 Stage 3 Dynamic Analysis

With the collection of static analysis, it is already possible to identify certain charac-
teristics of the malware, which leads us to the next phase that is dynamic analysis. This
type of analysis is to execute the sample to demonstrate its behavior and what capa-
bilities it has by being able to execute it. We can in some cases visualize what types of
changes are made in a computer, such as file modifications, file deletion, creation of
keys, among others.
Many of the tools that can be used for analysis are available on the Internet, there is
even a detailed guide called Practical Malware Analysis: The Hands-On Guide to
Dissecting Malicious Software, by the writer Michael Sikorski [8] which we will be
guided to know What tools to use.
2.4 Stage 4 Dynamic Analysis Support Tools

There is a considerable amount of online tools to support our analysis, for this reason
after the manual study of the sample it is possible to rely on the following websites that
have the ability to analyze malicious files [9].
• https://any.run/
• http://www.virustotal.com
• https://metadefender.opswat.com
• http://www.intezer.com/
• https://go.crowdstrike.com/try-falcon-prevent.html
• https://www.joesandbox.com/
• https://valkyrie.comodo.com
• https://www.reverse.it/
2.5 Stage 5 Report

Finally, all background related to the analysis must be used to prepare the corre-
sponding final documentation, or if necessary, use this background for another more in-
depth analysis phase, such as executing a reverse engineering process, that is, the study
of a sample in order to determine how it was built, disassembling its code to be able to
understand what it can do or what damage they cause, in short, to discover what an
object is made of, sample, software, etc. through conjectures that arise in the study.
3 Applied Example
3.1 Setting up a Work Environment

In order to work safely, it is necessary to configure a laboratory that has certain
restrictions to avoid damaging other computational assets, this is possible thanks to
virtualization software [6], which allow us to install and configure different operating
systems to perform analysis, but when configuring them it is necessary to consider at

least the following:
• Before working make a snapshot or clone of the machine.
• Once the sample is loaded on the machine, disable USB to prevent infections that
use this technology to spread.
• Once the sample is loaded on the machine, disable any network card in the device.
• Not having or creating shared folders on the network, there should be no connection
to the host computer.
• It is very likely that depending on the tools or capacity of the scripting malware that
needs to be analyzed, it is necessary to use different operating systems such as
Windows or one based on GNU/Linux. Therefore, it is advisable to configure at
least 2 different laboratories with the same sample, in this way it is possible to use
different tools or functionalities of each environment.
3.2 Stage 1 Identification of a Sample

We begin by identifying the sample that we will analyze, in this case a Java Script
called “hmskerx.js”. We will use the Kali Linux distribution and the “md5sum” tool to
perform an MD5 hash of the sample, using the following command (Fig. 2):
Fig. 2. Sample MD5.
Once the MD5 hash is obtained, we proceed to extract the metadata contained in
the sample, this may vary with each case (Fig. 3):
Fig. 3. Sample metadata.
When extracting the metadata and realizing that the type of file is unknown, we use
the “File” utility that comes by default in Kali Linux to obtain more data, as shown
below (Fig. 4):
Fig. 4. File type
This gives us an indication that the sample has ASCII characters and that it is also
composed of large lines of code.
3.3 Stage 2 Static Analysis of a Sample

In order to work with the information that the sample possesses, it is important to be
able to extract the hash that it contains in all its formats, for them we can use the
“hashcalc” tool and calculate not only the MD5 of this sample, this information could
be useful When you need to search through open sources for information on the sample
under analysis (Fig. 5).
Fig. 5. File type.
If we open an editor to analyze the sample, we come across what appears to be

obfuscation of the code, this to prevent antivirus detects the threat and can run without
problems, this is also possible to observe using the tool “strings” that shows us
information regarding the characters that can be printable per console as evidenced in
the following image (Fig. 6):
Fig. 6. Sample printable characters.

If we go to the end of the result we can observe the following (Fig. 7):
Fig. 7. Sample printable characters.
We already have some data regarding how the obfuscation of this script works,
which by the way uses WScript, an environment developed by Windows for the
execution of scripts in its operating system.
Now, with the background information obtained, it is possible to perform searches
in open sources, using tools such as “total virus” to analyze, for example, some of the
hashes of the sample, as it is possible to observe in the following image:
3.4 Stage 3: Dynamic Analysis of a Sample

3.4.1 Evidence Taking
In order to be able to show or understand the changes that a computer undergoes when
executing malicious script, it is necessary to identify what are the normal processes that
must be running before executing the sample (Fig. 8).
Fig. 8. Process evidence taking.

3.4.2 Sample Execution, Review and Process Comparison

At this point it is necessary to identify what processes were created after execution, it is
also important to visually inspect if there is any change in the environment such as
unknown extensions, file deletion, file duplication, etc. (Fig. 9).
Fig. 9. Taking evidence of new processes generated.
A good activity now is to review each of the processes generated to identify their
properties and where they are stored.
3.4.3 Comparison
To compare what changes exist in a machine once infected, a tool called “Regshot” can
be used which makes a comparison of the before and after a sample is executed, as can
be seen in Fig. 10, a “shot” to a clean machine, then the sample is executed, then a
second shot” is taken and finally the results must be compared.
This will give us a text file with all the changes that arose in the equipment once the
sample was executed, being able to show how many values were deleted, modified, as
well as folders or keys within the system.
3.4.4 Traffic Analysis

In order to know what kind of requests the sample makes, it is possible to use the
“ApateDNS” tool which helps answering DNS requests through the malicious script
without connecting to the Internet, and as with these tools it is possible to use
“Wireshark” to analyze the traffic and see what requests or requests are made on the
network and determine any malicious domain that tries to download some malicious
software once connected to the Internet.
Fig. 10. Tool for comparing changes in the system logs.
3.5 Stage 4 and 5: Dynamic Analysis Support Tools and Report

It is also possible to rely on online tools such as “Valkyrie” (https://valkyrie.comodo.
com) or “Any run” (https://any.run/) which are platforms that allow you to load
samples to be analyzed automatically and manual depending on the interest of the
analyst and once obtained the greatest amount of background the analyst must process
the information and draw from it a conclusion regarding the study, the report must be
ordered and in phases as proposed in this investigation.
4 Model Proposal
Based on the above, the following malware scripting analysis methodology is proposed
(Fig. 11):
Fig. 11. Proposal methodology on based 5 stage.

This methodology aims to facilitate the analysis of various types of scripts which may
contain some type of malicious use, considering that it is a quick study and must be
executed before conducting an in-depth analysis, in order to save time in making
decisions regarding an infection that could affect the normal development of activities
both in an organization and in the personal life of an analyst.
In addition, it is important to mention that this methodology is the basis for a deeper
research, leaving as future work to adapt new tools and lessons learned from other
disciplines [10], such as computer forensics, in terms of custody and recognition of
evidence. Thus, this paper is intended as a first approach to the issue at hand, in order to
provide the foundations for a thorough methodology proposal, tested against real-life
use cases.
References
1. Gutmann, P.: The commercial malware industry. In: DEFCON Conference (2007)
2. Baltimore, B.: Baltimore’s out-of-date and underfunded IT system was ripe for ransomware
attack (2019)
3. Baltimorebrew classification homepage. https://www.baltimorebrew.com/2019/05/21/
baltimores-out-of-date-and-underfunded-it-system-was-ripe-for-a-ransomware-attack/.
Accessed 21 May 2019
4. Gandotra, E., et al.: Malware analysis and classification: a survey. J. Inf. Secur. 5, 56–64
(2014). https://doi.org/10.4236/jis.2014.52006
5. Mieres, J.: Ataques informáticos. Debilidades de seguridad comúnmente explotadas.
Recuperado (2009). http://proton.ucting.udg.mx/tutorial/hackers/hacking.pdf
6. Mitnick, K.: The Art of Invisibility: The World’s Most Famous Hacker Teaches You How to
be Safe in the Age of Big Brother and Big Data. Hachette, UK (2017)
7. Gómez, J., Villar, E.: Introducción a la virtualización (2018)
8. Fedorenko, E., Ivanova, A., Dhamala, R., Bers, M.U. The Language of Programming: A
Cognitive Perspective. Trends in cognitive sciences (2019)
9. Sikorski, M., Honig, A.: Practical Malware Analysis: The Hands-On Guide to Dissecting
Malicious Software (2012)
10. Pandey, S., Mehtre, B.: Performance of malware detection tools: a comparison. In: 2014
IEEE International Conference on Advanced Communications, Control and Computing
Technologies, pp. 1811–1817. IEEE (2012)
11. Uppal, D., Mehra, V., Verma, V.: Basic survey on malware analysis, tools and techniques.
Int. J. Comput. Sci. Appl. (IJCSA) 4(1), 103 (2014)
Validation of ICS Vulnerability Related
to TCP/IP Protocol Implementation
in Allen-Bradley Compact Logix PLC
Controller
Jaime Pavesi1(&), Thamara Villegas1, Alexey Perepechko2,

Eleazar Aguirre3, and Lorena Galeazzi4
1
Facultad de Ciencias, Universidad Mayor, Santiago, Chile
{jaime.pavesi,thamara.villegas}@umayor.cl
2
Independent IOT/IIOT Security Researcher and Security Architect,
Amsterdam, The Netherlands
alexey.perepechko@gmail.com
3
eaguirre@cic.ipn.mx
4
Centro de Investigación en Ciberseguridad CICS, Universidad Mayor,
Santiago, Chile
lorena.galeazzi@mayor.cl
Abstract. Industrial Control Systems (ICS) research and testing process was
implemented to validate the existence of a well known security vulnerability in a
Rockwell Automation Allen-Bradley Compact Logix PLC controller. The study
was conducted considering a public advisory from the manufacturer, which
includes a large list of families of affected products by the vulnerability. The
established hypothesis of the study considered the existence of the vulnerability
in a specific available PLC model, included by Rockwell Automation manu-
facturer in the list of affected products. An exploit was developed and multiple
testing was performed to trigger the vulnerability.
Testing methodology and results indicates there is sufficient evidence to
establish that Rockwell Automation Allen-Bradley Compact Logix 5370 L2
controllers, are not affected by the same type of Improper Input Validation
vulnerability, than the Compact Logix 5370 L3 controllers, as it was stated by
the manufacturer in a public advisory.
Keywords: ICS PLC TCP/IP Security Vulnerability Exploit
1 Introduction
Industrial Control Systems (ICS) play a very important role in world economy and
current way of living. ICS became highly automated and network connected, intro-
ducing benefits but also new security vulnerabilities. This is particularly important in
Critical Infrastructure ICS such as energy, water, transportation, etc. [12].
Automation components of ICS include sensors, actuators, PLCs and SCADA
systems, among others components [8]. New ICS vulnerabilities are discovered every
https://doi.org/10.1007/978-3-030-33229-7_30
356 J. Pavesi et al.
year. In several countries, regulation forces manufacturers to recognize and inform ICS
vulnerabilities in public advisories, which usually include mitigation procedures and
recommendations to their customers, such as software or firmware upgrades, ICS
network isolation, traffic block, renew equipment, etc.
Such mitigations actions may have an important cost for the companies and are also
time consuming. Most of the companies rely on the information provided by ICS world
class manufacturers, such as Rockwell Automation, Siemens, etc.
When an ICS vulnerability become public, including a list of ICS affected products,
companies currently using those products feels they are at risk of being attacked. This
is particularly important in the case of critical infrastructure companies, whose oper-
ation depends on ICS systems that are supposed to be vulnerable [9].
Those companies don’t hesitate to follow the risk mitigation recommendations of
their provider/manufacturer, assuming the involved costs. But, what if the information
provided by manufacturers, related to ICS vulnerabilities, is not enough accurate?
What if ICS manufacturers decide to include families of products supposedly
affected by a vulnerability, in their public advisories, to prevent a possible damage,
without having complete demonstrated confirmation of the vulnerability, in some of the
products?
Looking for an answer to all these questions, this study focuses on technical val-
idation of PLC vulnerability in Rockwell Automation Allen-Bradley Compact
Logix PLC affected products, declared in NCICC Advisory ICSA-18-172-02 [1].
That specific PLC was selected for this research, because the research team has full
access to a Lab with 4 of those PLCs and also has access to the PLC configuration and
communication licensed software, which are needed for the purpose of the study.
2 Problem Definition and Hypothesis
In June 2018, researcher Alexey Perepechko from Applied Risk company based in the
Netherlands, discovered a PLC vulnerability and sent an Advisory Report [2] to
Rockwell Automation manufacturer. The report include the following main
information:
“A vulnerability in the Rockwell Automation Allen-Bradley CompactLogix 5370
Controller 1769-L30ERMS could allow an unauthenticated, remote threat actor to
reboot the device and switch the device to the “Major Non-Recoverable Fault” mode,
resulting in a Denial of Service (“DoS”) condition” [2].
“Vulnerability Details. The vulnerability is due to incorrect processing of TCP ACK
packet additional options by the listener at Ethernet/IP TCP port (default 44818). An
incorrect order on the NOP option leads to a immediate device reboot and enters a
“Major Fault” mode which must be resolved manually. To trigger the vulnerability, the
NOP option must be put first and the number of options must be more than one” [2].
The vulnerability discovered and reported, was recognized by the PLC manufac-
turer. Rockwell Automation determined there were additional products affected by the
vulnerability and reported this information to NCCIC [3], which result in NCICC
Advisory ICSA-18-172-02 [1]. List of affected products by the vulnerability, according
to Rockwell Automation [1, 13]:
Validation of ICS Vulnerability Related to TCP/IP Protocol Implementation 357
1. Allen-Bradley CompactLogix 5370 L1 controllers, Versions 30.014 and prior,

4. Allen-Bradley Armor CompactLogix 5370 L3 controllers, Versions 30.014 and
prior,
5. Allen-Bradley Compact GuardLogix 5370 controllers, Versions 30.014 and prior,
and
6. Allen-Bradley Armor Compact GuardLogix 5370 controllers, Versions 30.014 and
prior.
CVE-2017-9312 [14] and CWE-20 “Improper Input Validation” were assigned to
this vulnerability by NIST [4, 15] and MITRE [5] organizations. The specific controller
already mentioned, used to discover this vulnerability, belong to family number 3 of
affected products in the above list.
Universidad Mayor in Santiago, Chile, has an Automation Lab equipped with
Rockwell Automation controllers included in the list of affected products. The con-
trollers belong to family number 2 of affected products in the above list. The specific
PLC model is CompactLogix 5370 L24ER-QB1B. Based on the indicated vulnerability
information, a research was initiated with the following hypothesis.
Hypothesis: “Rockwell Automation Allen-Bradley CompactLogix 5370 L2 con-
trollers, have the Improper Input Validation vulnerability as stated by the manufacturer.
The vulnerability can be exploited, resulting in a Denial of Service condition of the
PLC”.
3 Methodology Design
The methodology to prove the stated hypothesis was based on the “vulnerability
details” contained in Sect. 2 of this paper, plus the analysis and definition of all the
elements that had to be considered. Several components needed to be defined:
• Confirmation of available controller with affected firmware version
• Confirmation of required conditions to trigger the vulnerability
• Define platform to perform testing
• TCP/IP protocol details study
• TCP packet Option Field details study
• TCP/IP network protocol analyzer
• TCP Parameters and Options for specific PLC model available
• Exploit method
• Restore procedure of the PLC system, in case of successful exploitation.
3.1 Confirmation of Available Controller with Affected Firmware

Version
Available Rockwell Automation PLC has firmware 21.01 and belongs to the list of
affected products, indicated in Sect. 2. All controllers with firmware version 30.014 or
prior are supposed to be affected, according to the manufacturer.
3.2 Confirmation of Required Conditions to Trigger the Vulnerability

As indicated in Sect. 2 in “Vulnerabilty Details”, the PLC must use Ethernet/IP pro-
tocol communication and have TCP port 44818 open (or other similar TCP port open).
Using NMAP scan tool, it was confirmed that TCP port 44818 is open in the PLC and
it´s used for Ethernet service. Therefore, these required conditions are ok (Fig. 1).
Fig. 1. Nmap scan confirms TCP port 44818 open for Ethernet-IP service
3.3 Defining a Platform to Perform the Tests

The Automation Lab at Universidad Mayor has an SMC FMS-200 manufacture plant,
with 4 PLCs connected to Ethernet network and 2 Desktops running Controller
Communication and Configuration software (Logix Designer software) over Windows
10 (Fig. 2).
Fig. 2. SMC FMS-200 manufacture plant with 4 PLCs
Logix Designer is already running on Windows in the Automation Lab. Logix

Designer was used to get the appropriate communication parameters with the PLC.
Research for available TCP/IP packets injection tools on windows was done, as it
will be explained in section “Exploit Method” of this paper.
The selected platform for TCP/IP packets injection was Linux. Specific platform
was Kali Linux version 2018.2 over VMWare virtual machine workstation version 15,
running over Windows 10, in a desktop connected to the PLCs network.
3.4 TCP/IP Protocol Details Study

TCP protocol is defined in RFC 793 [6] and IP protocol is defined in RFC 791 [7].
One of the main important concepts to understand regarding this protocol is the 3
Way Handshake (3WHS), which is used to establish a connection and also to finish a
connection.
3WHS consist in a sequence of three specific packets between a client and a server.
To establish a connection, the correct sequence is:
1. SYN packet
2. SYN + ACK packet
3. ACK packet
In this study, the Kali Linux environment is the Client and the PLC is the Server or
Host (Fig. 3).
Fig. 3. 3WHS: 3 Way HandShake connection sequence
The above figure allow to understand the 3WHS sequence needed to establish a
connection. It’s also important to notice that proper handle of sequence and
acknowledge numbering in TCP protocol, is required.
3.5 TCP Packet Options Field Details Study

TCP packet format must comply with RFC 793. Proper manipulation of TCP Options
field is key for the testing to trigger/exploit the known vulnerability, due to the vul-
nerability details already explained in Sect. 2 (Fig. 4).
3.6 TCP/IP Network Protocol Analyzer

To have network traffic visibility during the exploit testing process and to capture the
appropriate TCP parameters for valid communication with the PLC, it is mandatory to
have a network protocol analyzer tool. Wireshark tool was the preferred choice for this
purpose. It runs on Windows and Linux.
Fig. 4. Valid TCP packet format with multiple TCP options field
3.7 TCP Parameters and Options for Specific PLC Model Available
Each Rockwell Automation PLC model has specific communication parameters over
TCP/IP network protocol, with the communication and configuration software Logix
Designer. Those same parameters were used in the exploit code, to establish
Fig. 5. Wireshark capture showing valid communication parameters with the PLC
communication with the PLC. Wireshark was used to capture valid 3WHS traffic
between the PLC and Logix Designer (Fig. 5).
Therefore, for Compact Logix 5370 L24ER-QB1B specific available controller
model, the valid parameters and options are:
• Window = 2300
• Maximum Segment Size (MSS) option = 1460
• Window Scale (WS) option = 2
• Selective ACK ok (SAok) option = Yes
• Timestamp option = none.
3.8 Exploit Method

The options were basically two: finding a suitable and available TCP/IP packets
injection tool for Windows or Linux, or a customized software development. Several
tools were tried: Packet Sender, NetScanTools Pro, Trex, OpenStego, Nmap, Warp.
None of the above tools is capable to properly handling the Options Field of a TCP
packet, for suitable ACK and SYN injections. Therefore, the alternative was a software
development, which require to select a proper software language. After research, the
main language options were two:
1. C++ with Libnet library, a specific library for network applications
2. Python with Scapy library, also a specific library for network applications
Python with Scapy library was selected, because that was the same language and
library used by researcher Alexey Perepechko to successfully exploit the vulnerability
in the other PLC model, the Rockwell Automation Allen-Bradley Compact Logix 5370
L30ERMS.
3.9 Restore Procedure of the PLC System, in Case of Successful

Exploitation
Before starting the exploit testing, it was confirmed that in case of a major fault and a
Denial of Service condition of the PLC, we were capable of restoring the PLC to
normal operation, by reloading a backup of the firmware and the application software,
with the Logix Designer tool provided by the manufacturer.
4 Experimental Testing and Results
The testing consisted on running multiple exploit versions, while monitoring and
capturing network traffic with Wireshark. Valid parameters and options previously
shown were used, for Compact Logix 5370 L24ER-QB1B PLC model.
A total of 90 tests were done, changing the python exploit code to achieve the
following combinations:
• SYN crafted packet with 1 NOP option first and then MSS, WS and SAok options
• SYN crafted packet with 2 NOP option first and then MSS, WS and SAok options
• SYN crafted packet with 1 NOP option first and 1 NOP between MSS, WS, SAok
• Same previous one, but changing the location of the 2nd NOP in the Options
• ACK crafted packet with 1 NOP option first and then MSS, WS and SAok options
• ACK crafted packet with 2 NOP option first and then MSS, WS and SAok options
• ACK crafted packet with 1 NOP option first and 1 NOP between MSS, WS, SAok
• SYN & ACK crafted packets with 1 NOP option first and then MSS, WS and SAok
options
• SYN & ACK crafted packets with 2 NOP option first and then MSS, WS and SAok
options
• SYN & ACK crafted packets with 1 NOP option first and 1 NOP between other
options
• Additional tests, suppressing a RESET (RST) packet that was automatically sent by
Linux Kernel to the PLC, after receiving a SYN + ACK packet from the PLC
• Tests ended using the exact same python exploit code, with valid parameters, that
was capable of exploiting the vulnerability in the Compact Logix 5370
L30ERMS PLC model [10, 11].
In all these tests, communication with the Compact Logix 5370 L24ER-QB1B PLC
was established and confirmed, always having the 3 packets of the 3 Way Handshake,
always getting the SYN + ACK reply from the PLC. Multiple Wireshark network
traffic captures were saved for documentation purpose.
However, the Improper Input Validation vulnerability was never triggered for this
PLC model. There was no major fault of the controller and no denial of service
condition.
5 Conclusions
It´s important to notice that Alexey Perepechko, the same researcher that discovered
and exploited the vulnerability in the Rockwell Automation Allen-Bradley Compact
Logix 5370 L30ERMS PLC model, participated in this research to trigger/exploit the
same vulnerability in the Compact Logix 5370 L24ER-QB1B PLC model. Both PLC
models are supposed to be affected products by the tested vulnerability, according to
Rockwell Automation manufacturer.
Multiple exploit tests were done, including the exact same python exploit code,
with valid parameters, that was used in successful exploitation when this vulnerability
was discovered. Considering the experimental testing and results shown in Sect. 4, the
hypothesis of this research was false.
However, there is sufficient test evidence to establish that Allen-Bradley Compact
Logix 5370 L2 controllers, are not affected by the same type of Improper Input Val-
idation vulnerability, than the Compact Logix 5370 L3 controllers. This evidence is
opposed to the statement of the manufacturer, which informed NCICC that both
controllers are affected products by the same CWE-20 Improper Input Validation
vulnerability.
References
1. ICS CERT US Homepage. https://ics-cert.us-cert.gov/advisories/ICSA-18-172-02. Accessed
15 June 2019
2. APPLIED RISK Homepage. https://www.applied-risk.com/resources/ar-2018-002. Acces-
sed 15 June 2019
3. NCICC Homepage. https://www.us-cert.gov/. Accessed 15 June 2019
4. NIST Homepage. https://nvd.nist.gov/vuln/detail/CVE-2017-9312. Accessed 15 June 2019
5. MITRE Homepage. http://cwe.mitre.org/data/definitions/20.html. Accessed 15 June 2019
6. RFC Homepage. https://www.rfc-editor.org/info/rfc793. Accessed 15 June 2019
7. RFC Homepage. https://www.rfc-editor.org/info/rfc791. Accessed 15 June 2019
8. Serhane, A., Raad, M. Raad, R., Susilo, W.: PLC Code-level vulnerabilities. In: 3rd
International Conference on Computer and Applications (ICCA), Beirut, Lebanon, pp 348–
352. IEEE (2018)
9. Lee, R., Assante, M., Conway, T.: Analysis of the cyber attack on the Ukrainian power grid.
Technical report, E-ISAC, Washington, DC (2016)
10. Echeverri, D.: Hacking with Python. Zeroxword Computing, Madrid (2015)
11. Echeverri, D.: Python for Pentesters. Zeroxword Computing, Madrid (2014)
12. Krotofil, M., Gollmann, D.: Industrial control systems security - what is happening? In: 11th
International Conference on Industrial Informatics (INDIN), Bochum, Germany, pp 664–
669. IEEE (2013)
13. Milinkovic, S., Lazic, Ljubomir.: Industrial PLC security issues. In: 20th Telecommunica-
tions Forum (TELFOR), Belgrade, Serbia, pp. 1536–1539. IEEE (2012)
14. CVE Details Homepage. https://www.cvedetails.com/cve/CVE-2016-9343. Accessed 15
June 2019
15. NIST Homepage. https://nvd.nist.gov/vuln/detail/CVE-2016-9343. Accessed 15 June 2019
Identifying Components Belonging to Wireless
Connectivity Security
Lorena Galeazzi Avalos1(&), Cristian Barría Huidobro1,

and Thamara Villegas Berbesi2
1
Vicerrectoría de Investigación, Centro de Investigación en Ciberseguridad
CICS, Universidad Mayor, 7500628 Santiago, Chile
{lorena.galeazzi,cristian.barria}@mayor.cl
2
Vicerrectoría de Investigación, Facultad de Ciencias, Escuela Ingeniería
Electrónica, Universidad Mayor, 7500628 Santiago, Chile
thamara.villegas@umayor.cl
Abstract. Information systems security has been a need for people and
enterprises. This paper presents a review about the different variables studied in
the literature affecting normal operations in Wi-Fi, these variables can change or
affect the normal scenario, conditions, even vulnerabilities, hindering or facili-
tating attack by a third party. According to variables number identified, they are
grouped and related into four categories components. All of this in support of to
achieve a higher level of physical and information security.
Keywords: Components security wireless Communications devices

Transmition system Security of the environmental Variable
1 Introduction
Currently, data security has an important role for enterprises and people. Different
institutions have been trying to unify standards for risks evaluation linked to infor-
mation handling, creating criteria to guarantee data security, integrity and confiden-
tiality. This paper delivers base analysis to establish security level for WiFi networks,
focused on the way the different instructions are performed by the different programs
algorithms and their interaction and, the process management required to coordinate the
different components needed to build the communication up.
The study objective is identifying components belonging to wireless connectivity
security considering the variables that directly connect the physic and data link layers
from OSI model or the access layer from the TCP/IP model (see Fig. 1).

https://doi.org/10.1007/978-3-030-33229-7_31
366 L. Galeazzi Avalos et al.
Fig. 1. A figure OSI model, physical and data link layers (Source: prepared)
2 Development
A wireless network, as defined by 802.11x standard, radiofrequency technology, is

used as an alternative of using wired cable. This investigation is based on the wireless
transmission media and the identification of the variables from the different compo-
nents involved in the Wi-Fi security according to OSI and TCP/IP models, specifically
in the first layers [1].
Wireless networks are defined by the radio waves modulation and data transmitting
signaling [2]. Because of their nature, wireless networks are exposed and easily
identified or monitored. Hence, importance to have clearly identified the variables and
components belonging to the wireless network.
Components are catalogued according to the element characteristics, behavior and
effects on the Information System. Variables are the component elements that change
and impact the risk test results. Different components and associated variables can be
identified (see Fig. 2).
Identifying Components Belonging to Wireless Connectivity Security 367
Fig. 2. Variables grouped by components (Source: prepared)
2.1 Security
This component groups the variables in the protection level (see Fig. 3).
Fig. 3. Security component (Source: prepared)

The “User Knowledge Level”, refers to the knowledge level the user has to detect a
security fault or to configure the system to avoid the security issue. Users with
knowledge are able to apply security and use support tools like software and apps.
There are three types of “Password”, according to user level: device, technical
administrator and web administrator.
The variable “Router Security Feature”, as there are different types of devices, some
of them with defense incorporated for some kind of attacks and are considered a
security part and the first filter for any suspicious activity.
“Encryption Protocol”, are a first security barrier as cypher the data transmission
through WEP, WPA, WPA2 and soon WPA3.
Because of device variety and different kind of attacks available, the “Firewall” is a
containment barrier considered a variable. In case of an attack, can mitigate attacker
chances and increase router security.
The “Radius Server” is accountable of network access security. It controls if the
user and password are valid, gives access to servers to the user that is attempting to
access the wireless network.
2.2 Transmission System

Cataloged as a component because the elements amount related directly with the
transmission process. In the case of wireless networks, this transmission type corre-
sponds to a non-guided medium, as the electromagnetic waves are transmitted without
combination through the air. The radiofrequency signal are electromagnetic waves that
transport information [3].
The variables or elements part of this component are related to (see Fig. 4).
Fig. 4. Security transmission system (Source: prepared)

Effective “bandwidth” is the frequency band where is concentrated the most part
of the signal, which is limited by the transmission speed. Therefore, there is a direct
relation between transmission speed and bandwidth: the higher the transmission speed
of a signal, the greater the effective bandwidth [4].
The “attenuation” described as factors, like distance and atmospheric, are con-
ditions affect the transmission energy level and consequently the receiver errors level.
“Noise Signals” are present in the transmission media, non-wanted signals affect
between the transmitter and receiver, distorting the information signal. The noise can
be classified in four categories: Thermal noise, Intermodulation noise, Crosstalk and
Impulsive noise.
“Signal Loss” is consequence of different effects [5], like atmospheric absorption,
and absorption through the interaction with the matter like walls, glass, concrete,
marble, wood, stone, etc. that can be defined as signal absorbers.
The “Physics Effects”, such as diffraction, is energy redistribution in a wave front
when goes at the border of an object; allows wave propagation through the corners [6].
Reflection, when the electromagnetic wave pass through the border between two
mediums, a part of the wave is reflected in the opposite direction [6]. Refraction,
changes the signal direction by passing in an oblique direction from a medium to
another, with different propagation velocity [6].
The “Distance”, defined as the range of space or time that mediates between two
things or events [7]. In wireless transmissions, the signal is dispersed by distance,
therefore the greater the distance from the transmitter, the transmission power range
will be considerably reduced [8].
2.3 Communication Devices

These are the devices needed for a wireless network connection and operation (see
Fig. 5).
“Antenna” is an electric conductor used for transmitting or receiving electromag-
netic signals from bi-directional communications. The antenna converts the electrical
energy from the transmitter to electromagnetic energy and then is radiated to the
atmosphere, space or water [4]. By the other side, the antenna converts the electro-
magnetic signal to electric energy and is sent to the receiver [4].
The antennas are designed to transmit and receive in a specific frequency [9].
Antennas are designed considering distance to reach, radiation intensity, beam width
and power. The radiation intensity, is the three dimensional radiation pattern, and
shows the direction the antenna radiates the signal [9]. This investigation only will treat
the antennas used for transmitting and receiving Wi-Fi wireless networks. The antennas
are classified as omnidirectional and directional, the most of devices used in wireless
networks use omnidirectional antennas. Directional antennas are used mainly for long
distance connection or to cover a specific area.
The variable “Network Interface Card” is an essential part when penetration test
tools are used [10]. Network interface cards should have the monitor mode enabled to
make the penetration test. Network interface cards have a chipset used for penetration
testing, but not all chipsets allow monitor mode, therefore, is important to know which
chipset [12] can be enabled to monitor mode [11].
Fig. 5. Communication devices (Source: prepared)
This variable considers the differences that wireless device manufacturer, in this
case described like “Vendor”, have according to their policies, production schemes,
support and updates. hat the first paragraph of a section or subsection is not indented.
According to the vendor, it is the degree of security at the administration level and
protection barriers that can be estimated as preliminary data to consider, at the time of
performing a pentesting.
Updates are a characteristic associated with the manufacturer, it defines whether the
device is vulnerable or not. It is also possible to mention, the location installed device,
normally they are not installed according to what the installation manual says, It may
not be installed as recommended by the manufacturer’s manual, for ignorance or
changes made without considering the initial implementation, This is an advantage for
any attacker, since the power varies and change the emitted signal direction by the
transmitter. Facilitating detection attacker activity, where the attacker can detect a
signal wireless networks, from any vulnerable place for any of penetration attack.
To pentesting level process, the amount “Connected wireless clients” to the devi-
ces, it is considered a variable. Mainly in the process attack, there are different types,
depend of traffic users generated when navigate on the network, for be used by a
software exploitation execution.
The Operative System of the endpoint, wireless router, and any tools of pentesting
software, are considered as an important variable for vulnerability level. The config-
uration systems and applications installed by operative systems, influence bad practices
adopted by users and administrators, for simple carelessness.
In process of the activities to be carried out, the “Equipment for Penetration Test”,
it is considered like a variable, considering different stages or phases of process, such as
recognition, attack, among others. The activities depend on different devices such as a
cell phone, notebook or tablet, among others. An interesting example is a notebook or
mobile used from a car in motion to detect wireless networks (wardriving) [13].
2.4 Environmental Safety

Internal and external environment where a wireless network is operating defines the
security requirements and vulnerability options. The access level in the neighborhoods
as well as the security personal presence can modify this component. That is the reason
to consider following variables (see Fig. 6).
Fig. 6. Safety of the environmental (Source: prepared)
The “Physical Security” is determined by the physical difficulties of location and

infrastructure of the unit, important features, to decide the level of network security by
the user and by the attacker to make a malicious activity.
People presence around the site to difficulting access attacker, determined like
variable “Personal security”, which directly affects on any activity to be carried out,
both in the process of pentesting, as the physical security network.
“Access security” is determined as the set of physical elements that allow you to
control access to the network. These elements are related to technology, focused on the
function of the company or organization, in support of control and information

security.
The first paragraphs that follows a table, figure, equation etc. does not have an
indent, either.
The “coverage” variable is related to the previous ones, allows the preparation and
use of masking tools, any type of activity associated with violating security without
being detected. Coverage is the way to penetrate the place with high security without
being detected.
This is directly related to conditions analysis, the way in which the safety tests will
be carried out is considered, and the conditions of access to information, two models
can be defined like “white box” or “black boxes” models.
The white box refers to the test procedures that use information on internal per-
formance and the source code in the case of transparent software testing, this type of
test has a focus on the information that the pentesting assessor or performer has, both at
the systems level and at the site access [13].
The black box refers to the tests that are carried out without knowledge of the
internal network infrastructure, its internal resources or the dependencies of the place to
be tested, without access permissions at the software and physical level of the place.
Therefore, pentesting is executed in a condition of unauthorized [13].
The black box refers to the tests that are carried out without knowledge of the
internal infrastructure of the network, its internal resources or the dependencies of the
place to be tested, without access permissions at the software and physical level of the
place. Therefore, pentesting is executed in a condition of unauthorized [13].
Its internal resources or the dependencies of the place to be tested, without access
permissions at the software and physical level of the place. Therefore, pentesting is
executed in a condition of unauthorized [13].
The regulations, regulations and “laws” in force in the country, and the internal
security procedures carried out by the company or person in its network must be
considered.
3 Conclusion
Identified components have associated the variables that affect the wireless transmis-
sion medium. Some of these variables are not considered in other studies or have been
considered on an isolated way. As result of this paper, these variables have been
grouped to find a relationship between them and define a real scenario affecting the
pentesting procedures. Penetration tests fail to achieve their objectives as only use ideal
environment parameters leaving aside other key parameters.
Therefore, each component variables compilation offers a new transmission med-
ium vision, providing decision making support to feasible attacks and make wireless
network security improvements.
This investigation defines future works on a guide design to evaluate wireless
network security level according to identified components.
References
1. International Organization for Standardization (ISO homepage) (2019). https://www.iso.org/
standard/20269.html
2. Pacio, G.: Data Center hoy. Buenos Aires - Argentina, año, Editorial Alfaomega (2014)
3. Stallings, W.: Comunicaciones Redes de Computadores. 7ma Edición (2004)
4. Hi, Y., Boile, K.: Antennas from Theory to Practice. Wiley, United Kingdom (2008)
5. Morán, J.: Efectos de interferencia sobre redes basadas en el estándar IEEE 802.11 en el
espectro de 2.4 GHz (2012)
6. Eleftheriades, G.V., Balmain, K.G.: Negative Refraction Metamaterials: Fundamental
Principles and Applications. Wiley-IEEE Press, Hoboken (2005)
7. https://dle.rae.es/?id=Dy4u4UW
8. Barclay, L.W.: Propagation of Radio Wave. IEE (2003)
9. Beard, C.: Wireless Communication Networks and Systems, William Stallings. Pearson,
London (2016)
10. Joshua, W., Cache, J.: Hacking Exposed Wireless: Wireless Security Secrets & Solutions.
McGraw-Hill Osborne Media, New York (2014)
11. Aircrack-ng Homepage, June 2019. https://www.aircrack-ng.org/doku.php?id=
compatibility_drivers
12. Ron, S., et al.: Analytic Methods in Systems and Software Testing. Institute, Technion, Israel
Fabrizio Ruggeri CNR-IMATI, Italy Frederick W. Faltin The Faltin Group and Virginia
Tech, USA (2018)
13. Hurley, C., et al.: WarDriving & Wireless penetration Testing. Syngress Publishing, Inc.
Rockland, Canada (2007)
Author Index
Acosta-Bermejo, Raúl 247 Gómez-Buenfil, R. A. 294

Aguilar Sierra, Verónica 3 Gómez-Buenfil, Romeo Alam 284
Aguirre, Eleazar 355 González León, Jacobo Gerardo 145
Aguirre-Anaya, Eleazar 247 Gordillo-Ramírez, A. 183
Aldana-Bobadilla, Edwin 268 Guzmán-Tolosa, José Misael 284
Alonso-Cuevas, O. 183 Guzmán-Toloza, J. M. 294
Álvarez-Carranza, Geraldine 25
Araujo Diaz, David 16 Hernández-Gómez, Jorge J. 45
Arellano-Verdejo, Javier 61 Hruby, Florian 3
Arjona-Villicaña, P. David 335 Hurtado Moreno, Juan José 171
Barria Huidobro, Cristian 213, 325 Lazcano-Hernández, Hugo E. 25

Barría Huidobro, Cristian 345, 365 Lopez-Ramirez, Cristhian 162
Beltrán-García, Pamela 247 Loza Pacheco, Eduardo Eloy 192
Burbano Gonzalez, Clara Lucia 213, 230
Burbano Gonzalez, Diana Carolina 213 Martinez Nuño, Jesus Alfredo 16
Mata Rivera, Miguel Félix 145
Camacho Campos, Luis Angel 16 Mata, Félix 121
Carmona García, Enrique Alfonso 203 Medina, Isaac 45
Castro-Ynurreta, Iván Antonio 258 Mejía M., Julio E. 230
Catalan-Salgado, Edgar-Armando 162 Menchaca García, Felipe Rolando 305
Chadwick, Carreto Arellano 305 Mendoza Valencia, Juvenal 171
Cordero Vidal, David 325 Meneses González, Salvador Ricardo 110
Cordero, David 345 Miguel Santiago, David 85
Córdoba U., Andrés Felipe 230 Millán, Paul 121
Couder-Castañeda, Carlos 45 Molina-Puc, J. K. 294
Coutiño, Jorge Humberto Muñoa 3 Molina-Puc, Jeni 284
Mondragón, Margarita 34
De la Borbolla del Valle, Genghis 3 Montiel, Jorge Erik 305
De la Rosa, Hernán 34 Montilla M., Jessica María 230
Díaz Sosa, Mayra Lorena 192 Moreno Galván, Elizabeth 203
Moreno-Sabido, Mario Renán 45
Enciso-Aguilar, Mauro Alberto 45 Muñoz, Diego 345
Escamilla-Ambrosio, Ponciano Jorge 247
Esquivel-Mendiola, Edgar Adrian 268 Nieto Sánchez, Felipe de Jesús 171
Evelyn, Gutiérrez León 305 Novelo-Cruz, Raúl 45
Galeana-Zapién, Hiram 268 Ojeda-Cepeda, René Alberto 258

Galeazzi Avalos, Lorena 365 Orea-Flores, Izlian Y. 85
Galeazzi, Lorena 355 Orozco-del-Castillo, Mauricio Gabriel 45,
Garay-Jiménez, Laura I. 85 284
García, Blanca 130 Ortega-Pacheco, D. 183
376 Author Index
Pavesi, Jaime 355 Silva Hernández, Iliana Amabely 71

Peniche-Ruiz, L. J. 294 Silva-Trujillo, Alejandra G. 335
Peniche-Ruiz, Larissa Jeanette 258
Peraza Garzón, Álvaro 71 Torres Ruiz, Miguel Jesús 192
Perepechko, Alexey 355
Pool-Couoh, Edwin Moises 258
Vega García, Nayeli 130
Vega Pérez, Jaime 130
Ressl, Rainer 3
Vélez-Saldaña, U. 183
Rivero-Angeles, Mario E. 85
Villafaña-Gamboa, D. F. 294
Rodríguez Márquez, Rita Trinidad 110
Villafaña-Gamboa, Dakar Fernando 284
Rodríguez Paz, Oscar 3
Villegas Berbesi, Thamara 365
Rodríguez Zamora, René 71
Villegas, Thamara 355
Rodríguez-Morayta, M. J. 294
Salmerón-Quiroz, Bernardino Benito 34 Yañez-Casas, Gabriela Aurora 45

Sandoval-Reyes, Sergio 98
Sarabia-Osorio, Humberto 284 Zagal-Flores, Roberto 162

10.1007@978 3 030 33229 7 PDF

Uploaded by

Copyright:

Available Formats

10.1007@978 3 030 33229 7 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

10.1007@978 3 030 33229 7 PDF

Uploaded by

Copyright:

Available Formats

Miguel Felix Mata-Rivera

Communications in Computer and Information Science 1053

Telematics and Computing

Editorial Board Members

Cristian Barría-Huidobro (Eds.)

Telematics and Computing

ISSN 1865-0929 ISSN 1865-0937 (electronic)

© Springer Nature Switzerland AG 2019

Nowadays, interest in telematics is increasing, especially as the ﬁeld continues to be

November 2019 Miguel Felix Mata-Rivera

Workshop Chair in Spatial Intelligence

Informatics Security Track

Program Committee (Research Papers)

Sisi Zlatanova TU Delft, The Netherlands

José-Antonio León-Borges UQROO, Mexico

GIS and Climate Change

Real Geographies in Virtual Space: A Practical Workflow

Cross-Layer Architecture Applied to Geospatial Information

Methodology to Create Geospatial MODIS Dataset . . . . . . . . . . . . . . . . . . . 25

Data Acquisition System for Solar Panel Analysis. . . . . . . . . . . . . . . . . . . . 34

Pattern Recognition Through Empirical Mode Decomposition for

Moderate Resolution Imaging Spectroradiometer Products

A Dynamic System Model Using Agile Method Techniques

Telematics and Electronics

Statistical Properties of Vehicle Residence Times for Fog

12–18 GHz Microwave Frequency Band Microstrip Patch Antenna

Analytics for Basic Products in Mexico . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Design and Development of Photovoltaic Power Meter . . . . . . . . . . . . . . . . 130

Artificial Intelligence and Machine Learning

Clustering Methodology in Mixed Data Sets. . . . . . . . . . . . . . . . . . . . . . . . 145

American Sign Language Electromiographic Alphabet Sign Translator. . . . . . 162

Artificial Intelligence as a Competitive Advantage

Software Engineering and Education

Augmented Reality with Swift in ARkit and Their Applications

Computer Tool Prototype for the Selections of Views to Materialize

Intelligent Learning Ecosystem in M-Learning Systems . . . . . . . . . . . . . . . . 213

Application of Cyberphysical Systems Through Logistics as a Mediating

IoT Botnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

A Clustering-Based Approach to Base Station Assignment

Fuzzy Logic Controller for Automation of an Autonomous Irrigation

Internet of Things for Irrigation System . . . . . . . . . . . . . . . . . . . . . . . . . . . 294

Architecture Proposal for Low-Cost Hydroponic IoT Based Systems . . . . . . . 305

Learning and Automatization of Searching in Digital Media, Based

Methodology for Malware Scripting Analysis in Controlled Environments

Validation of ICS Vulnerability Related to TCP/IP Protocol Implementation

Identifying Components Belonging to Wireless Connectivity Security . . . . . . 365

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

Florian Hruby1,2(&) , Rainer Ressl1,

Keywords: Virtual reality Immersion Computer graphics Game engine

© Springer Nature Switzerland AG 2019

To create such an illusion, highly immersive VR relies on stereo head-mounted dis-

1.2 From GIS to Immersive VR Environments

Fig. 2. GIS base data used for the MXVR project

2.2 Object Modelling

Object-Based Modelling. The probably most genuine approach of rebuilding a

To achieve 2D images a 3D object can be reconstructed from, any commercial

3 Composition of the VR Environment

Fig. 6. GeoIVE before (left) and after (right) ambience setting.

3.2 Static Elements