Scribd
Upload
3 views6 pages

Detecting Lumpy Skin Disease in Cows

This document presents a deep learning framework for detecting Lumpy Skin Disease in cows, which is caused by the Neethling virus and leads to significant health issues in cattle. The authors developed a model using various convolutional neural networks, achieving an accuracy of 92.5% in classifying infected and normal cow images. The study highlights the importance of timely diagnosis and the potential of machine learning in veterinary medicine.

Uploaded by

6046 UMA SHANTHI. S
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
3 views6 pages

Detecting Lumpy Skin Disease in Cows

This document presents a deep learning framework for detecting Lumpy Skin Disease in cows, which is caused by the Neethling virus and leads to significant health issues in cattle. The authors developed a model using various convolutional neural networks, achieving an accuracy of 92.5% in classifying infected and normal cow images. The study highlights the importance of timely diagnosis and the potential of machine learning in veterinary medicine.

Uploaded by

6046 UMA SHANTHI. S
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

EasyChair Preprint

№ 2795

A Deep Learning Approach to Detect Lumpy Skin


Disease in Cows

Gaurav Rai, Naveen, Aquib Hussain, Amit Kumar and


Rahul Nijhawan

EasyChair preprints are intended for rapid


dissemination of research results and are
integrated with the rest of EasyChair.

February 28, 2020


A Deep Learning Approach to Detect Lumpy Skin Disease
in Cows

Gaurav Rai , Naveen , Aquib Hussain , Amit Kumar , Rahul Nijhawan

Abstract - Disease in animals is now a common [Link] disease is characterized by large skin nodules
problem. There are several types of disease in animals covering all parts of the body, fever ,nasal discharge, spread
so the identification of disease is important and lymph nodes and lachrymation. Lumpy skin disease is
diagnosis will be done at a timely pace. Lumpy skin generally found in Africa, Russia ,Egypt , Oman and India. At
disease in cows is caused by a virus called Neethling. first it was identified in Egypt. Less commonly, the virus may
By the affection of these diseases the cattles permanent be spread by direct contact to the skin lesions, saliva, nasal
damage to their skin. The disease often results in discharge, milk, or semen of infected animals [1].
reduce milk projection, infertility, poor growth, abortion Unfortunately, there are no specific antiviral drugs available
and sometimes death. We have developed an for the treatment of Lumpy Skin Disease. The only
architecture using machine learning techniques to treatment available is supportive care of cows. We can also
propose the disease or detect the disease. This think about treatment of skin lesions using wound care.
framework developed using tools like VGG-16 , VGG-19 Sprays and the use of antibiotics to prevent secondary skin
and Inception-v3 for extracting the features. The work is infections and Pneumonia [2].
tested on our dataset and is measured with other
advanced methodology kNN, SVM, NB, ANN and LR There are various parts of the paper in which Section 1 is
which results in considerable performance in extraction Introduction, Section 2 is Data Collection, Section 3 is
of features. proposed approach, Section 4 is results and Section 5 is
Conclusion.
Keywords - Feature Extraction, Dataset Collection,
Cow images infected from Lumpy Disease and Normal cow
Deep Learning , Lumpy Skin Disease, etc.
image (uninfected) are given by fig 1.

I. INTRODUCTION
Skin is an important part of an animal body. Lumpy skin
disease is a viral disease in cows extent by biting
.

Fig 1. Images of Lumpy Skin and Normal Skin in cows

Revised Manuscript Received on July 22, 2019. Naveen, CSE, Graphic Era University, Dehradun, India. Email:
* Correspondence Author ynaveen569@[Link]
Gaurav Rai, CSE, Gurukula Kangri Vishwavidyalaya, Haridwar, Aquib Hussain, CSE, Graphic Era University, Dehradun,
Uttarakhand, India. Email: gauravrai15011998@[Link] Uttarakhand, India. Email: aquibhussain738@[Link]
II. DATA COLLECTION Since the problem with the dataset is that it can be biased
The collection of dataset is an important phase because due to lack of proper dimensions and resolution but in this
there is no dataset available of lumpy skin disease in cows. dataset we have chosen only high resolution and proper
So we created an image dataset of cows around 300-400 pixel images [6].
images that is shown in table 1. Table. 1

Types of Cow Number of Images

Infected from Lumpy Disease 132

Normal Images of Cow 199

III. PROPOSED APPROACH


In this framework we used deep convolutional Neural is responsible for reducing the images dimensionality which
Networks. We take an image as input, process it and classify is used to overcome the problem of overfitting .The results of
it in various categories. In this process it consists of many the Pooling layer are passed to the Flattening Layer formed
layers in which the first layer is a convolution layer that as input layer, then we use the Artificial Neural Network [7].
operates the feature extraction by using filter or Kernel. The mathematical equation of CNN is given in fig 2.
Since the images are non-linear so to remove the linearity of
images we used

ReLU function(Rectifier Linear Unit) [3]. The result of the


convolution layer passed to the next layer, the Pooling layer

Fig 2.

16, VGG-19 and Inception-V3 for feature Extraction. We split


The workflow of the model is given in figure 3. In the the dataset into 75% as training data and rest 25% as testing
workflow Data Sampler is used to split the dataset and data [8].
Image Embedder uses different models such as VGG-
Fig 3. Workflow

We evaluate the ROC curve for the calculation of represents the ROC curve on Lumpy Skin Disease and Fig 5
performance on our framework. On X-Axis “Specificity” is represents the ROC curve on normal cow images [7
defined and on Y-Axis “Sensitivity” is mentioned. Fig 4

Fig 4. ROC curve on Lumpy skin disease

Fig 5. ROC curve on Normal cow images

IV. RESULTS

In this Section, we come up with the final result achieved by used for feature extraction and applies various classification
the tests done in the above section. In this project we have models like KNN , SVM, ANN, LR and NB , observing that
used three scenarios. In the first scenario Inception-v3 is we got the highest accuracy on test data of 92.5% with
Artificial Neural Network that is shown in Table 2. In the scenarios we observe the best model is Inception-v3 with
second scenario we used VGG-16 for feature extraction and Artificial Neural Network with highest accuracy 92.5%. The
got 87.9% accuracy with Logistic Regression. In the third accuracy Assessment table is given below.(Table 2)
scenario, a 19-layer convolutional Neural Network (VGG-19)
is used for extracting the features and got the 88.2%
accuracy with the Naive Bayes. By observing all the three
Table 2. Accuracy Assessment

V. CONCLUSION

In our algorithms we used Deep Convolutional Neural different fields of medical science like to detect skin cancer
Network to predict Lumpy skin and Normal Skin that are in animals and other diseases. Moreover, it can help
predicted from our given dataset which obtained result of veterinary surgeons to figure out animal disease problems in
92.5% accuracy. Since there is no standard dataset early stages that do not require much manual work.
available, there are many problems that can arise but we
used the standard quality of image dataset due to which our
model is working properly. This model can be used in

VI. ACKNOWLEDGEMENT

We would like to express our special thanks to Dr. Ankush


Mittal (300+ publications and 2 international Books) and Dr.
Rahul Nijhawan for the proper guidance and motivation at
various stages.

Reference -
1. Annandale, Cornelius & Holm, Dietmar & Ebersohn, 3. Shadab, Malik & Dwivedi, Mahavir & S N, Omkar & Javed,
Karen & Venter, Estelle. (2013). Seminal Transmission of Tahir & Bakey, Abdul & Raqib, Mohammad & Chakravarthy,
Lumpy Skin Disease Virus in Heifers. Transboundary and Akshay. (2019). Disease Recognition in Sugarcane Crop
emerging diseases. 61. 10.1111/tbed.12045. Using Deep Learning. 10.13140/RG.2.2.21849.47209.

2. Tuppurainen, Eeva. (2017). Epidemiology of Lumpy skin


disease.
4. Nijhawan, Rahul & Das, Josodhir. (2020). A Proposed Detection Using CNN based Feature Extraction. 1-7.
Framework Approach for Mapping Glacier Hazard Zones. 10.1109/ICECCT.2019.8869364.
10.1007/978-981-13-7067-0_44.

11. Nijhawan, Rahul & Rishi, Megha & Tiwari, Amit & Dua,
5. Nijhawan, Rahul & Jindal, Radhika & Sharma, Himanshu Rajat. (2019). A Novel Deep Learning Framework Approach
& Raman, Balasubramanian & Das, Josodhir. (2019). A for Natural Calamities Detection: Proceedings of Third
Deep Learning Framework Approach for Urban Area International Conference on ICTCS 2017.
Classification Using Remote Sensing Data. 10.1007/978-981-13-0586-3_55.
10.1007/978-981-32-9088-4_37.
6. Rawat, Sarthak & Bisht, Abhishek & Nijhawan, Rahul.
(2019). A Classifier Approach using Deep Learning for 12. Nijhawan, Rahul & Das, Josodhir & Raman,
Human Activity Recognition. 486-490. Balasubramanian. (2018). A hybrid of deep learning and
10.1109/ICIIP47207.2019.8985926. hand-crafted features based approach for snow cover
mapping. International Journal of Remote Sensing. 40. 1-15.
7. Aishwarya, & Goel, Akansha & Nijhawan, Rahul. (2019). A 10.1080/01431161.2018.1519277.
Deep Learning Approach for Classification of
Onychomycosis Nail Disease. 13. Nijhawan, Rahul & Das, Josodhir & Raman,
10.1007/978-3-030-30577-2_98. Balasubramanian. (2018). A Hybrid CNN + Random Forest
Approach to Delineate Debris Covered Glaciers Using Deep
Features. Journal of the Indian Society of Remote Sensing.
8. Chhabra, Harshit & Srivastava, Akshay & Nijhawan, 46. 1-9. 10.1007/s12524-018-0750-x.
Rahul. (2019). A Hybrid Deep Learning Approach for
Automatic Fish Classification.
10.1007/978-3-030-30577-2_37. 14. Nijhawan, Rahul & Raman, Balasubramanian & Das,
Josodhir. (2018). Proposed hybrid-classifier ensemble
algorithm to map snow cover area. Journal of Applied
9. Nijhawan, Rahul & Joshi, Deepankar & Narang, Naman & Remote Sensing. 12. 1. 10.1117/[Link].12.016003.
Mittal, Aditya & Mittal, Ankush. (2019). A Futuristic Deep
Learning Framework Approach for Land Use-Land Cover 15. Kozma, Robert & Ilin, Roman & Siegelmann, Hava.
Classification Using Remote Sensing Imagery. (2018). Evolution of Abstraction Across Layers in Deep
10.1007/978-981-13-0680-8_9. Learning Neural Networks. Procedia Computer Science.
144. 203-213. 10.1016/[Link].2018.10.520.

10. Varshni, Dimpy & Thakral, Kartik & Agarwal, Lucky &
Nijhawan, Rahul & Mittal, Ankush. (2019). Pneumonia
He did his internship from RWX Technology and successfully
completed his project “Mushroom Detection” using Machine
AUTHORS PROFILE Learning.

Gaurav Rai is currently pursuing [Link] (4th


year) in Computer Science & Engineering from
Gurukula Kangri Vishwavidyalaya, Haridwar,
Uttarakhand. He did his internship from RWX Aquib Hussain is computer science student
Technology and successfully completed his from Graphic Era University, Dehradun,
project “Breast Cancer Detection” using Machine Learning. Uttarakhand. He did his internship from RWX
Technology and successfully completed his
project on Machine Learning

Naveen is computer science student from


Graphic Era University, Dehradun, Uttarakhand.

Common questions

Powered by AI

The primary challenge in creating a dataset for diagnosing lumpy skin disease is the lack of existing datasets, as there are no standard datasets available for this specific disease . This necessitated the creation of a new image dataset, consisting of 300-400 high-resolution images to avoid bias and ensure proper image dimensions and pixel quality . Despite these measures, the absence of a standard dataset raises concerns about potential biases and the accuracy of model training and validation .

Using deep learning techniques for early disease detection in animals like cows has several positive implications. It allows for the timely identification and management of diseases, reducing the risk of severe outbreaks and potential economic losses due to decreased milk production, infertility, or death . Furthermore, these techniques minimize the need for manual intervention, enabling veterinary surgeons to address disease problems at early stages, potentially improving animal health outcomes and streamlining veterinary processes .

The deep convolutional neural network architecture used in this research includes components such as convolutional layers for feature extraction using filters, ReLU functions for introducing non-linearity, pooling layers for dimensionality reduction, and flattening layers for transitioning to an input layer suitable for artificial neural network processing . These components collectively enhance the model's ability to capture and process intricate details of image data, contributing to its high effectiveness and accuracy in detecting lumpy skin disease .

The ReLU (Rectifier Linear Unit) function assists in processing non-linear images by applying a non-linear activation to the outputs of a convolutional layer. This activation removes negative values, thereby allowing the model to account for and represent complex patterns more effectively. It helps introduce non-linearity to the model, making it capable of learning detailed and intricate features necessary for accurately detecting lumpy skin disease .

Biting insects play a significant role in spreading lumpy skin disease among cows by acting as vectors that carry the virus from infected to healthy animals. The disease primarily affects areas in Africa, Russia, Egypt, Oman, and India, indicating its prevalence in these regions due in part to the presence of insect vectors .

Distinctive symptoms of lumpy skin disease in cows include large skin nodules covering the body, fever, nasal discharge, swelling of lymph nodes, and lachrymation. These symptoms can lead to reduced milk production because the disease can cause physical stress and discomfort, impacting the cow's ability to produce milk efficiently .

Developing antiviral drugs for lumpy skin disease is challenging because the disease is caused by the Neethling virus, and there are currently no specific antiviral drugs available . Supportive treatments recommended include wound care for skin lesions, sprays, and antibiotics to prevent secondary skin infections and pneumonia, focusing on managing symptoms rather than curing the disease .

The deep learning model developed for lumpy skin disease can be adapted for other animal diseases by altering the training dataset to include relevant disease-specific images and modifying model parameters to adjust for the unique features of each disease. By selecting appropriate architectures for feature extraction, much like the use of Inception-v3 or VGG-16, the model can be calibrated to recognize patterns specific to new diseases, potentially aiding in early detection and treatment across various animal health contexts .

The Inception-v3 model contributes to high accuracy by efficiently performing feature extraction from images. This model, when paired with artificial neural networks, leverages its comprehensive architecture to enhance the detection of complex patterns associated with the disease. The combination achieved the highest accuracy of 92.5% in comparison to other models tested, highlighting the effectiveness of this approach .

Deep convolutional neural networks, specifically those utilizing Inception-v3 with artificial neural networks, outperformed other methods like kNN (K-Nearest Neighbors) and SVM (Support Vector Machine) in detecting lumpy skin disease. The VGG-16 and VGG-19 models achieved accuracy levels of 87.9% and 88.2%, respectively, with Logistic Regression and Naive Bayes, but the superior accuracy of 92.5% using a CNN with Inception-v3 and ANN highlights its enhanced feature extraction and classification capabilities .

You might also like