Implementation of Lumpy Skin Disease Detection

Using Machine Learning Approach

1st Prof. Dipali Himmatrao Patil 2nd Mahesh Pawar 3rd Mukesh Jaiswal
Department of Information Technology Department of Information Technology Department of Information Technology
JSPM’s Rajarshi Shahu College of JSPM’s Rajarshi Shahu College of JSPM’s Rajarshi Shahu College of
Engineering, Pune-411033, India Engineering, Pune-411033, India Engineering, Pune-411033, India
patil.dipali07@gmail.com maheshpawar20july@gmail.com mukesh.jaiswal2k1@gmail.com

5th Sandesh Jagtap

4th Prasad Rane Department of Information Technology
Department of Information Technology JSPM’s Rajarshi Shahu College of
JSPM’s Rajarshi Shahu College of Engineering, Pune-411033, India
Engineering, Pune-411033, India sandeshjagtap1441@gmail.com
prasadkrane21@gmail.com

Abstract— Lumpy skin disease (LSD), a contagious and health of animals. Machine learning and image
ailment that can swiftly disseminate within cattle recognition techniques have shown promise in
populations, presents a serious risk to the cattle recent years for the identification and diagnosis of a
industry's finances. Timely identification and precise
number of illnesses, including LSD in cattle. Lumpy
recognition of afflicted animals are imperative to
Skin Disease, characterized by its highly infectious
impede the transmission of the infection and mitigate
its impact. Cattle and buffalo are the primary hosts of nature and its ability to affect cattle, poses a
this viral malady, underscoring the urgency of effective significant financial burden on the livestock sector.
measures. It is important to note that there is no health
danger associated with LSD, however, it can seriously
Lumpy Skin Disease, characterized by its highly
impair the welfare of animals and reduce output. infectious nature and its ability to affect cattle, poses
Images are preprocessed using Convolutional Neural a significant financial burden on the livestock sector.
Networks (CNN) in order to extract features. The next
step we will take is to use machine learning models The lumpy skin disease virus (LSDV) belongs to the
such as Inception V3. Poxviridae family, more precisely to the genus
Capripoxvirus and the subfamily Chordopoxviridae.
Our research is on the application of several machine This illness usually affects creatures that are related
learning algorithms for the diagnosis of lumpy skin to cows, such as water buffalo and cattle. The
disease in cattle. We have used total five machine
sickness varies in severity according on the species.
learning models and calculated performance metrics
for each model to determine how well models can
Fever, depression, nasal discharge, skin nodules,
work. Consequently, Inception V3 outperformed the enlarged lymph nodes, and tears are typical
other models, achieving 97.4% accuracy and symptoms. Reduced milk supply, problems with
98.1% precision. fertilization, delayed development, and perhaps
higher rates of illness and death are all possible
Keywords: Lumpy skin disease, Inception V3, outcomes of the condition.
LSD, Neural Networks, Mobile Application.
Numerous issues may be resolved with machine
learning approaches, such as regression, clustering,
I. INTRODUCTION and classification. Algorithms used in machine
learning include reinforcement learning, supervised
The livestock business suffers significant financial learning, and unsupervised learning[5]. These
losses due to the extremely infectious viral illness techniques are frequently used to solve problems
known as lumpy skin disease (LSD), which can pertaining to categorization. Cattle illness is
affect cattle. Early identification and diagnosis of classified and detected using a variety of well-liked
affected animals is essential for stopping the spread machine learning classification techniques, such as
of illness and lessening its effects on the production
neural networks, Inception V3, VGG, AlexNet, and Additive exPlanations) visualizations to gain more
ResNet 101. insight into the accuracy of the classification, since
Random Forest performs well with the chosen
A deep learning algorithm, or convolutional neural dataset.
network (CNN), is a subset of machine learning.
This method belongs to the category of artificial The researchers in this study [3] employed SMOTE
neural networks and may be used to many kinds of as a resampling approach and Random Forest
data and applications. A CNN is a type of network classification to maximize the prediction of Lumpy
architecture/algorithm that processes pixel input and Skin Disease. They also utilized a genetic algorithm
is used for image recognition. The design of a CNN for hyperparameter tuning. Recall measures are
is comparable to the structure of brain connection. employed in this work to classify LSD, with the
notion being that the greater the score, the higher the
The CNN uses three different layers. The essential probability that the system will label cows with the
part of a CNN is its convolutional layers, which are infection wrongly as healthy.
where elements like edges, textures, and forms are
extracted from the input picture by applying filters. The author employs a number of machine learning
After the convolutional layers' output is processed models in this study [4]. The purpose of this
through pooling layers, the feature maps are down- investigation was to determine whether or not cattle
sampled to lower the spatial dimensions while were afflicted with lumpy skin disease. A total of ten
keeping the most crucial data. One or more fully machine learning classifiers were utilized, and
connected layers are then applied to the output of the assessment metrics were calculated to determine the
pooling layers in order to categorize or forecast the efficacy of each classifier. The two classifiers that
picture. have performed the best overall are the Random
Forest Classifier and the Light Gradient Boosted
Convolutional neural networks (CNNs) with object Machine Classifier.
identification and picture categorization are used in
the Inception V3 architecture. Researchers at In this article [5], author focuses on forecasting
Google invented it. Computer vision applications whether or not cattle in a given geographical area are
including image recognition, object detection, and susceptible to Lumpy disease in the future in order
picture segmentation commonly employ Inception to implement preventive measure. They discovered
V3. that Random Forest, Adaboost, and Neural Network
can reliably forecast the probability of LSDV
infection based on several studies.
II. LITERATURE REVIEW

The author of this report [1] proposes a deep learning III. SOFTWARES AND TOOLS
model based on convolutional neural networks
(CNNs) to accurately detect a variety of illnesses VS Code Editor
from photographs, including infectious bovine
keratoconjunctivitis (IBK), lumpy skin disease It's a versatile and strong editor. VS Code is a robust
(LSD), and foot and mouth disease (FMD). Early and versatile code editor due to its many features and
detection of diseases decreases human error, and extensions. has fundamental qualities. With features
make diagnosis available to all farmers. This might like live servers, snippets, git integration,
completely transform the way disease is controlled. extensions, and a debugger, Visual Studio Code is a
Although further study is required to perfect the great tool for web development.
model and incorporate it into farm management Keras Neural Network API
systems, there is a great deal of promise for
enhancing cattle health and reducing financial Python-based Keras is a deep learning API that may
losses. be used with PyTorch, TensorFlow, or Java
augmented reality. It is written in Python. An
Using robust Machine Learning algorithms based on effective Application Programming Interface (API)
geographic and meteorological data, the author's for advanced neural networks is called Keras.
approach [2] for detecting Lumpy Skin Disease Designed to make deep neural network research
Virus (LSDV) is based on developing a prediction more rapid, this open-source neural network
model. Random Forest outperforms other framework may be used with CNTK, TensorFlow,
algorithms in terms of predicting the likelihood of and Theano. Keras focuses on being modular, user-
LSDV prediction. We also employ SHAP (SHapley friendly, and extensible. Keras runs on top of
TensorFlow and is relatively fast. In Keras, learning framework. Custom layers can be created
prototyping takes less time. and added to a model to introduce specific
functionalities or features tailored to a particular task
Google Colab or application.
Google Colab, or Collaboratory, is a cloud-based Create and compile the Model:
platform tailored for collaborative coding, with a
specific emphasis on machine learning and data Creating and compiling a neural network model
analysis. Python code notebooks may be created by involves defining the architecture of the model,
users on Google's cloud platform, where they can specifying how it should be trained, and configuring
have free access to powerful processing resources parameters such as the optimizer, loss function, and
such as CPU, GPU, and TPU. This enables the metrics.
execution of sophisticated machine learning
algorithms and efficient handling of substantial Sigmoid activation Function
datasets. Noteworthy for its user-friendly sharing This is also called the transfer function used for
features, seamless integration with Google Drive, getting the output of the node. Activation functions
Google Colab finds popularity in educational are used to determine the output of the neural
environments, research endeavors, and the network like yes or no. The resulting values of
development of machine learning projects.
activation functions are between 0 to 1 or -1 to 1.
The range of sigmoid is between 0 to 1. This is an S-
shape curve. The sigmoid function is
IV. MATERIAL AND METHODOLOGY defined by the formula:

A. Data Set Sigmoid Function:

This study used a dataset of cattle that are infected

1
S(x) = where e is an Euler's number and
through the lumpy disease and the normal cattle. Our 1−𝑒 −𝑥
dataset, comprises more than 2000 cattle images. S(x) is the sigmoid function (2.718).
That are classified into three categories Healthy,
Mild, and Severe. The collection contains pictures of
cattle taken throughout various LSD stages as well Data Preparation:
as pictures of livestock taken without LSD. The
dataset image size is 256 X 256 and An essential first step in training machine learning
the colors is RGB. models is data preparation, especially in the context
of computer vision where images need to be
B. Methodology preprocessed before being fed into the model.
Resizing images to a specified target size, such as
Import Libraries: 299x299 pixels, is a common preprocessing step.
In Python, importing libraries allows you to access
pre-built functions and tools to enhance your code.
The import statement is used to bring in external
libraries.

Load Inception V3 Model:

A deep convolutional neural network (CNN)

architecture, the InceptionV3 model is a member of
the Inception model family. It's an evolution of the
original Inception model, also known as
GoogLeNet. InceptionV3 is designed for image
classification and computer vision tasks.

Add Custom Layers:

Adding custom layers refers to incorporating layers

into a neural network architecture that are not part of
the standard predefined layers provided by a deep
Figure. 1: Data Preparation Model
Train the model:

Training a deep learning model is a fundamental

process that involves optimizing the model's
parameters to minimize a chosen objective function,
commonly known as the loss functionThis function
provides a quantitative assessment of the model's
performance on the training data by analyzing the
difference between the model's predictions and the
actual target values.

Application Model:

Figure 4: Homepage

In Figure 4, the homepage of the proposed model is

displayed, offering "Sign Up with Email" for new
users and "Login" for existing users.

Figure 2: Application Model

C. Working of an application

Figure 5: Disease Information

Figure 5, outlines the Overview of Lumpy Skin

Disease, designed to enhance awareness. It
provides information on symptoms, causes, risk
factors, and prevention. Users can choose an image
from the device/internal storage for testing, and the
last result is stored.

Figure 3: Image Capturing

Figure 3 Illustrates the procedure for accessing the

Android mobile app after downloading the .zip file.
 output for sharing with doctors is available, enabling
users to save the result internally or to the cloud.

D. RESULTS

As demonstrated in (Table 1) and visualized in

(Graph 1), the suggested application provides an
average accuracy of 97.40% for predicting the
proper outcome from the additional image. About
1940 photos make up the dataset used for detection,
which is divided into 3 groups to train the
algorithm.

Comparative Analysis:
Algorithm Accuracy Precision Recall F1

Inception 97.4 98.1 96.7 97.4

V3
Alex-Net 93.8 95.1 95.1 93.8

ResNet101 87.2 91.2 87.2 88.5

Random 89.7 94.5 90.0 92.2

Forest
VGG 16 82.1 88.2 82.1 83.9

Figure 6: Alert Message

Figure 6, presents a cautionary message, alerting Table 1: This table represents comparative analysis
users to the model's 91% accuracy and suggesting of performance measures of different CNN models
prompt consultation with a doctor.
100
97.4
95 93.8
89.7
90
87.2
85 82.1

80

75

70
Inception Alex-Net ResNet 101 Random VGG 16
V3 Forest
Accuracy

Figure 8: This graph represents accuracy of different

CNN Models.

LUMPY SKIN DISEASE DETECTION USING OUR

MODEL:

Confusion Matrix:
Figure 7: Result Message
The prediction summary is represented as a
Figure 7, showcases an image of a tested cow for matrix in a confusion matrix. It displays the
LSD. The result indicates whether the cattle is number of correct and wrong predictions for
normal, mild, or severe, along with the accuracy rate each class. It aids in comprehending the
for that specific image. The option to download the
courses that are perplexed by the model as .
another class.
Figure 11: Loss Graph of our model using
Inception V3 with increasing no of images.

Actual Values
Predicted 550(TP) 10 (FP) VI. CONCLUSION
Values 5 (FN) 5 (TN)
The primary goal of our model is to predict the
Table 2: Confusion Matrix
prevalence and severity of bovine TB while
minimising financial losses in the cattle industry.
Performance Metrics Our findings emphasise the need of collecting high-
quality data sets from numerous sources in order to
100
increase machine learning model performance, as
98 98.1
97.4 well as how machine learning and image recognition
96.7
approaches may be utilised to identify LSD in cattle.
96 Our programme has the potential to enhance the
94.7
94 quick identification and diagnosis of LSD in cattle
with an accuracy of 97.4%, which might assist to
92 curtail the disease's spread and alleviate the related
Accuracy Precision Recall F1 Score economic losses.

Figure 9: Performance Metrics of Inception V3 VII. REFERENCES

[1] M. Rony, D. Barai, Riad and Z. Hasan, "Cattle External Disease
Classification Using Deep Learning Techniques," 2021 12th
International Conference on Computing Communication and
Networking Technologies (ICCCNT), Kharagpur, India, 2021, pp. 1-
7, doi: 10.1109/ICCCNT51525.2021.9579662.

[2] A. K. Jain, R. G. Tiwari, N. Ujjwal and A. Singh, "An Explainable

Machine Learning Model For Lumpy Skin Disease Occurrence
Detection," 2022 International Conference on Data Analytics for
Business and Industry (ICDABI), Sakhir, Bahrain, 2022, pp. 503-508,
doi: 10.1109/ICDABI56818.2022.10041665.

[3] Suparyati, E. Utami and A. H. Muhammad, "Lumpy Skin Disease

Prediction Based on Meteorological and Geospatial Features using
Random Forest Algorithm with Hyperparameter Tuning," 2022 5th
International Conference on Information and Communications
Technology (ICOIACT), Yogyakarta, Indonesia, 2022, pp. 99-104,
doi: 10.1109/ICOIACT55506.2022.9971807.

[4] D. F. Dofadar, H. M. Abdullah, R. H. Khan, R. Rahman and M. S.

Ahmed, "A Comparative Analysis of Lumpy Skin Disease Prediction
Figure 10: Accuracy Graph of our model using Through Machine Learning Approaches," 2022 IEEE International
Conference on Artificial Intelligence in Engineering and Technology
Inception V3 with increasing no of images (IICAIET), Kota Kinabalu, Malaysia, 2022, pp. 1-4, doi:
10.1109/IICAIET55139.2022.9936742.

[5] N. Ujjwal, A. Singh, A. K. Jain and R. G. Tiwari, "Exploiting Machine

Learning for Lumpy Skin Disease Occurrence Detection," 2022 10th
International Conference on Reliability, Infocom Technologies and
Optimization (Trends and Future Directions) (ICRITO), Noida, India,
2022, pp. 1-6, doi: 10.1109/ICRITO56286.2022.9964656.

[6] D. Dommeti, S. R. Nallapati, C. Lokesh, S. P. Bhuvanesh, V. Vara

Prasad Padyala and P. V V S Srinivas, "Deep Learning Based Lumpy
Skin Disease (LSD) Detection," 2023 3rd International Conference on
Smart Data Intelligence (ICSMDI), Trichy, India, 2023, pp. 457-465,
doi: 10.1109/ICSMDI57622.2023.00087.

[7] V. Narayan et al., "Severity of Lumpy Disease detection based on

Deep Learning Technique," 2023 International Conference on
Disruptive Technologies (ICDT), Greater Noida, India, 2023, pp. 507-
512, doi: 10.1109/ICDT57929.2023.10150925.
 Global Journal of Computer Science and Technology, pp.
13–18, 2019.
[8] S. Gambhir, S. Khanna and P. Malhotra, "Machine Learning Based [20] dansbecker, “Rectified linear units (ReLU) in deep
Diagnosis of Lumpy Skin Disease," 2023 International Conference on learning,” Kaggle.com, 07-May-2018. [Online]. Available:
Artificial Intelligence and Applications (ICAIA) Alliance Technology https://www.kaggle.com/dansbecker/rectified-linear-units-
Conference (ATCON-1), Bangalore, India, 2023, pp. 1-5, doi: relu-indeep-learning. [Accessed: 04-Jun-2021].
10.1109/ICAIA57370.2023.10169125.
[21] B. Ramsundar and R. B. Zadeh, TensorFlow for Deep
[9] A. Kumar, B. Kumar and H. S. Negi, "Predicting Lumpy Skin Disease Learning. Sebastopol, CA: O’Reilly Media, 2018.
using Various Machine Learning Models," 2023 International
Conference on Computational Intelligence and Sustainable [22] “Multi-Class Neural Networks: Softmax,” Google.com.
Engineering Solutions (CISES), Greater Noida, India, 2023, pp. 412- [Online]. Available:
416, doi: 10.1109/CISES58720.2023.10183604. https://developers.google.com/machine-
learning/crashcourse/multi-class-neural-networks/softmax.
[10] M. Rony, D. Barai, Riad and Z. Hasan, "Cattle External Disease [Accessed: 04-Jun2021].
Classification Using Deep Learning Techniques," 2021 12th
International Conference on Computing Communication and [23] R. Karim, “Illustrated: 10 CNN Architectures,” Towards
Networking Technologies (ICCCNT), Kharagpur, India, 2021, pp. 1- Data Science, 29-Jul-2019. [Online]. Available:
7, doi: 10.1109/ICCCNT51525.2021.9579662. https://towardsdatascience.com/illustrated-10-cnn-
architectures95d78ace614d. [Accessed: 06-Jun-2021].
[11] M. K. Ojha, D. Reji Karakattil, A. D. Sharma and S. Mary Bency,
"Skin Disease Detection and Classification," 2022 IEEE India Council [24] Harris, I., Osborn, T. J., Jones, P., & Lister, D. (2020).
International Subsections Conference (INDISCON), Bhubaneswar, Version 4 of the CRU TS monthly high-resolution gridded
India, 2022, pp. 1-5, doi: 10.1109/INDISCON54605.2022.9862834. multivariate climate dataset. Scientifc data, 7(1), 1-18.
https://doi.org/10.6084/m9. fgshare.11980500
[12] A. Ajith, V. Goel, P. Vazirani and M. M. Roja, "Digital dermatology:
Skin disease detection model using image processing," 2017 [25] Gilbert, M., Nicolas, G., Cinardi, G., Van Boeckel, T. P.,
International Conference on Intelligent Computing and Control Vanwambeke, S. O., Wint, G. W., & Robinson, T. P.
Systems (ICICCS), Madurai, India, 2017, pp. 168-173, doi: (2018). Global distribution data for cattle, bufaloes, horses,
10.1109/ICCONS.2017.8250703. sheep, goats, pigs, chickens and ducks in 2010. Scientifc
data, 5(1), 1-11. https://doi.org/10.1038/sdata.2018.227
[13] R. G., S. G. A., K. S. and R. R., "Skin Disease Detection based on
Machine Learning Techniques," 2022 11th International Conference [26] Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer,
on System Modeling & Advancement in Research Trends (SMART), W. P. (2011). Smote: synthetic minority over-sampling
Moradabad, India, 2022, pp. 1249-1252, doi: technique. arXiv eprints. arXiv preprint arXiv:1106.1813.
10.1109/SMART55829.2022.10047634.

[27] Bharat Singh Thakur, “Analyzing A Cattle Health

[14] K. V. Swamy and B. Divya, "Skin Disease Classification using Monitoring System Using Iot And Its Challenges In Smart
Machine Learning Algorithms," 2021 2nd International Conference on Agriculture”, EPRA International Journal of
Communication, Computing and Industry 4.0 (C2I4), Bangalore, Multidisciplinary Research.
India, 2021, pp. 1-5, doi: 10.1109/C2I454156.2021.9689338.

[28] Noone Vijay Kishan, Sai Trinath Y, Sandeep Kavalur,

Sangamesha V, Mr. Sumanth Reddy, “Cattle disease
[15] K. Roy, S. S. Chaudhuri, S. Ghosh, S. K. Dutta, P. Chakraborty and identification using Prediction Techniques”,
R. Sarkar, "Skin Disease detection based on different Segmentation IJARIIEISSN(O)-2395-4396.
Techniques," 2019 International Conference on Opto-Electronics and
Applied Optics (Optronix), Kolkata, India, 2019, pp. 1-5, doi:
10.1109/OPTRONIX.2019.8862403.

[16] S. A. Alanazi et al., “Boosting breast cancer detection using

convolutional neural network,” J. Healthc. Eng., vol. 2021,
p. 5528622, 2021.

[17] Y. Sun, L. Zhu, G. Wang, and F. Zhao, “Multi-input

convolutional neural network for flower grading,” J. Electr.
Comput. Eng., vol. 2017, pp. 1–8, 2017.

[18] J. Brownlee, “How do convolutional layers work in deep

learning neural networks?,” Machinelearningmastery.com,
16-Apr-2019. [Online]. Available:
https://machinelearningmastery.com/convolutional-layers-
for-deeplearning-neural-networks/. [Accessed: 04-Jun-
2021].

[19] M. A. Hossain and M. S. Alam Sajib, “Classification of

Image using Convolutional Neural Network (CNN),”