Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 60 views

    Remote Sense Big Data Analytics

    Uploaded by

    Ghassani Herning Pradita
    This document discusses using deep learning techniques for remote sensing big data analysis. It first describes the increasing complexity, diversity and dimensionality of remote sensing data. It then discusses challenges of large, data-intensive remote sensing tasks. Opportunities from improved computing platforms and data management tools are also presented. The document outlines common remote sensing data processing workflows and demonstrates applications for food security monitoring in Indonesia including rice classification and yield estimation. It analyzes results from applying deep learning models to remote sensing datasets and discusses proposals for improving deep learning methods for satellite imagery analysis.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Remote Sense Big Data Analytics

    Uploaded by

    Ghassani Herning Pradita
    60 views30 pages
    This document discusses using deep learning techniques for remote sensing big data analysis. It first describes the increasing complexity, diversity and dimensionality of remote sensing data. It then discusses challenges of large, data-intensive remote sensing tasks. Opportunities from improved computing platforms and data management tools are also presented. The document outlines common remote sensing data processing workflows and demonstrates applications for food security monitoring in Indonesia including rice classification and yield estimation. It analyzes results from applying deep learning models to remote sensing datasets and discusses proposals for improving deep learning methods for satellite imagery analysis.

    Original Description:

    Presentation about Remote Sense (Big Data Analytics from Paddy Land Satellite Photograph)

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses using deep learning techniques for remote sensing big data analysis. It first describes the increasing complexity, diversity and dimensionality of remote sensing data. It then discusses challenges of large, data-intensive remote sensing tasks. Opportunities from improved computing platforms and data management tools are also presented. The document outlines common remote sensing data processing workflows and demonstrates applications for food security monitoring in Indonesia including rice classification and yield estimation. It analyzes results from applying deep learning models to remote sensing datasets and discusses proposals for improving deep learning methods for satellite imagery analysis.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    60 views30 pages

    Remote Sense Big Data Analytics

    Uploaded by

    Ghassani Herning Pradita
    This document discusses using deep learning techniques for remote sensing big data analysis. It first describes the increasing complexity, diversity and dimensionality of remote sensing data. It then discusses challenges of large, data-intensive remote sensing tasks. Opportunities from improved computing platforms and data management tools are also presented. The document outlines common remote sensing data processing workflows and demonstrates applications for food security monitoring in Indonesia including rice classification and yield estimation. It analyzes results from applying deep learning models to remote sensing datasets and discusses proposals for improving deep learning methods for satellite imagery analysis.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 30

    Deep Learning for Big Data

    in Remote Sensing
    Mohamad Ivan Fanany, Sidik Mulyono, T. Basaruddin

    Faculty of Computer Science Universitas Indonesia


    in cooperation with Indonesian Agency
    of Assessment and Application of Technology (BPPT)

    Page 1

    High Resolution Earth Observation

    Increasing Complexity, Diversity, Dimensionality,


    Page 2

    Remote Sensing Big Data


    Challenges
    Increasing complexity of RS data: diversity and higher
    dimensionality.
    Large data-intensive issues: tasks and data
    dependencies
    Current platform are CPU-heavy but I/O- poor which do
    not work non-contiguously.

    Page 3

    Remote Sensing Big Data


    Opportunities from Tech. Leapfrogging
    1. Some supercomputers and Cloud computing
    platforms optimized for data-intensive loads;
    2. Parallel file systems and databases takes
    the data availability and locality as the main
    concern;
    3. The data managing tools for memory data
    placement controlling for multilevel data
    locality;
    4. Task scheduling focusing on large amount of
    dependent tasks and data availability.
    Page 4

    Remote Sensing Data Processing Flow

    Page 5

    Remote Sensing Utilization for


    Indonesian Food Security Program
    Rice field area: 8.1 million hectares
    Population: 240 million
    National rice need: 32 million T/year

    Page 6

    Paddy Growth Stages Classification


    Hyperspectral HyMap Image

    SAMPLING AREA: Karawang, West Java Indonesia


    (06 15 46.24S, 107 25 05.1E)

    Page 7

    Rice Yield Estimation

    The composited Images from 10


    lines of HyMap Hyperspectral
    Airborne Campaign
    (Karawang 2011)

    Paddy yield distribution map


    with GA-NSPCR
    Total production: 186,898 Ton
    Harvested Area: 30,966 Ha
    Page
    Yield average: 6.00 Ton/Ha

    Remote Sensing Data Fusion for


    Integrated Prediction

    HyMap Hyperspectral
    (4.4 m spatial res
    124 band spectral res)

    MODIS Multispectral
    (1.000 m spatial res
    7 band spctral res)

    Hyperspectral based
    MODIS integration model
    Yield
    prediction

    Growth
    stages

    Page 9

    Overall accuracy
    = 84.96%
    Statistical Validation

    Page 10

    Recent Results
    after Cloud Detection
    Recent
    and Removal Only on MODIS Data

    Page 11

    Recent Results
    after Cloud Detection
    Recent
    and Removal Only on MODIS Data

    Page 12

    Recent Results
    after Cloud Removal
    Recent
    Only on MODIS Data

    Page 13

    Scaling Up Model Building


    Beyond Supersite Limited Estimation

    Page 14

    Deep Learning for Accurate Land


    Cover and Land Use Prediction

    Page 15

    Deep Learning for Accurate Land


    Cover and Land Use Prediction

    Page 16

    Deep Learning for Accurate Land


    Cover and Land Use Prediction

    Page 17

    Analysis on DeepSAT Datasets


    (Basu et al. 2015)

    Hand Crafted Features as Deep Learning Input,


    Page 18

    Analysis on DeepSAT Datasets


    (Basu et al. 2015)
    Classification Accuracy using DBN,

    Conventional DL

    Proposed DL
    Page 19

    Analysis on DeepSAT Datasets


    (Basu et al. 2015)

    Classification Accuracy using CNN,


    Page 20

    Analysis on DeepSAT Datasets


    (Basu et al. 2015)

    Classification Accuracy using SDAE


    Page 21

    Analysis on DeepSAT Datasets


    (Basu et al. 2015)

    Classification Accuracy using SDAE


    Page 22

    Why Traditional Deep Architectures


    are Not Enough for SAT Datasets?
    Satellite datasets have higher intra and inter-class
    variability.
    The amount of labeled data is much smaller as
    compared to the total size of the dataset.
    Higher-order texture features are a very important
    discriminative parameter for various land cover
    classes.
    Shape/edge based features are not very useful in
    learning data representations for satellite imagery.
    Spatially contextual information is another important
    parameter fo.r modeling satellite imagery.
    Page 23

    Satelite Image Feature Ranking

    Page 24

    What are we currently doing?


    Proposal by Tanaka et al. 2013

    Page 25

    What are we currently doing?


    Proposal by Tanaka et al. 2013

    Page 26

    What are we currently doing?


    Proposal by Tanaka et al. 2013

    Page 27

    What are we currently doing?


    Proposal by Tanaka et al. 2013

    Page 28

    What we are currently doing?


    Some Lingering Questions
    How far Deep Learning can go?:
    Texture based prediction of vegetation and its
    growth stages
    Flatten frequency bands of MODIS or even
    Hyperspectral data into a single channel for
    classification?
    Are commonly used remote sensing parameters
    such as NDVI, EVI, GPP in multi or hyper bands
    based classification have higher Distribution
    Separation Criterion?
    Page 29

    THANK YOU
    Q&A

    Page 30

    You might also like