Lumpy Disease Classification Using Deep Learning

HARSHITH M P K M SUMUKH KASHYAP KARTIK VIRUPAKSHAPPA RODAGAI

Department of Information Science and Department of Information Science and Department of Information Science and
Engineering Engineering E ngineering
Sapthagiri College of Engineering,India Sapthagiri College of Engineering,India Sapthagiri College of Engineering,India
[email protected] [email protected] [email protected]

KAEERTHAN KUMAR NAYAK Prof. GAYATHRI R

Department of Information Science and Department of Information Science and
Engineering Engineering
Sapthagiri College of Engineering,India Sapthagriri College of Engineering,India
[email protected] [email protected]

ABSTRACT Lumpy skin disease (LSD) is a major threat to animal health and agricultural productivity because of its high
contagiousness resulting from lumpy skin disease virus (LSDV). The results of this study suggest a new approach to the
detection of LSDV that may help meet the need for early LSD control. The proposed approach will utilize image processing
and IoT technologies for faster detection of lumpy skin diseases in cattle in order to facilitate timely intervention.
Technology is now an essential element in contemporary society, transforming people’s lives and promoting luxury. One
can see the transition through many projects such as Advanced Cattle Health Monitoring System using Arduino and IoT.
Critical parameters for cattle health such as body temperature, respiration, heartbeat and rumination are monitored
continuously within the wireless sensor-based cattle health monitoring system. While the ESP8266 wifi module ensures
clamor less communication between different node sensors within the framework. The Thingspeak app allows one to
visualize data trends so that informed decisions regarding livestock management are made. It is important to note that
this study emphasizes the possibility of using image processing and IoT technologies in LSDV diagnosis and animal
management improvement, respectively, in order to improve agricultural practices and enhance animal welfare.

KEY WORDS Adulteration detection, brewing process, deep learning, image processing, machine vision.

I. INTRODUCTION the animals' symptoms. In case of abnormal readings, the

As more and more people experience the joy of owning a system will automatically alert veterinarians or hospital staff,
pet, the number of animal hospitals continues to increase. Our enabling swift action that could potentially save lives. Our
furry friends bring us endless happiness, from being loyal research focuses on tracking and monitoring the heart rate and
companions to providing security and becoming cherished temperature of sick animals during crucial phases such as
members of our families. Of course, ensuring the health and post-treatment recovery or after surgical procedures. By
well-being of your pet is the most important thing for the providing caregivers with timely insights and alerts, our web
owner. But the biggest challenge for veterinarians and medical application aims to enhance the quality of care provided to
staff is the inability to provide regular care to animals. sick animals, ultimately improving their chances of recovery
Hospitals often do not have procedures to monitor and alert and overall well-being. In addition to tracking and monitoring,
people caring for animals at all hours of the day. This our web application will also provide analysis and reporting
difference in care puts animals at greater risk of disease, of sick animal data to veterinarians. Dairy cattle, being
makes it harder for them to recover, and leads to more sensitive creatures, require constant monitoring of vital signs
violence. Understanding animal health requires important like body temperature and heart rate. Maintaining optimal
information such as pulse, respiration, body temperature and conditions is essential for their health and productivity, as any
blood pressure. These data points act as lifelines in healthcare deviations can lead to various health issues. Moreover,
systems, alerting caregivers to any abnormalities that may environmental factors like humidity can also impact cattle's
signal danger to the animal's well-being. To bridge this gap, health and milk quality. High levels of stress due to
our research proposes the development of a userfriendly web environmental factors can reduce milk quality and jeopardize
application designed specifically to monitor and record vital overall herd health. By utilizing our technology, dairy
signs, starting with dogs and cats. The primary aim is to track ranchers can improve milk profitability and quality while
and monitor the heart rate and temperature of sick animals. ensuring the well-being of their herds, thus providing a higher
Any irregularities in heart rate can serve as early warning level of animal welfare and security.
signs, prompting caregivers to intervene promptly. Our web
application will enable continuous monitoring and recording II. LITERATURE SURVEY
of heart rates and temperatures, providing caregivers with A Real-time App for Detecting Bovine Disorders: This study
real-time insights into their patients' health. Additionally, the presented at the IC3IoT conference introduces a smart mobile
system will generate easy-to-understand diagrams illustrating app built on ConvNet technology. The app efficiently
the recorded data, making it simple for caregivers to analyze identifies common bovine diseases like Bovine Mastitis and
Lumpy Skin Disease using image classification. Achieving an leveraged the significant sources of the net to supplement
accuracy of 98.58%, this app offers farmers a valuable tool their series. By looking on line, they were able to gather extra
for early disease detection, ultimately enhancing cattle photos of farm animals with LSD.
farming practices.
2.Data Splitting and Labeling:
2.Deep Learning for Cattle Disease Classification: Published The researchers compiled a total of 250 images. To make
in the ICCCNT conference, this paper proposes a certain the model's effectiveness at some stage in schooling
model to detect prevalent cattle external diseases using deep and checking out, they meticulously divided the dataset.
learning techniques like CNNs. By utilizing various CNN Eighty percent (2 hundred pictures) had been detailed for the
architectures, the model achieves 95% accuracy, reducing schooling section. This larger component allows the version
human errors in disease identification and aiding veterinarians to analyze and refine its potential to distinguish between
and farmers in prompt disease recognition. healthful and LSD-infected farm animals. The last 20% (50
pictures) have been set apart for trying out. This testing
3.IoT for Disease Detection in Cows: dataset serves as a benchmark to assess the model's accuracy
Highlighting the healthcare challenges in the dairy industry, in actual-world eventualities.
this paper presents an IoT-based approach for detecting Foot
and Mouth Disease (FMD) and Mastitis in cows. By 3.Data Preprocessing for Enhanced Quality:
employing sensors and machine learning algorithms, this Prior to feeding the images into the model, the researchers
system facilitates early disease detection, crucial for undertook a information preprocessing step. This important
minimizing economic losses and ensuring cattle health. level aimed to enhance the excellent of the snap shots and
eliminate any noise or inconsistencies that might preclude the
4.Image Processing and ML for Lumpy Skin Disease version's overall performance. Techniques like grayscale
Detection: conversion, median filtering, and comparison enhancement
Focused on Ethiopia's livestock sector, this research tackles had been employed to gain this. Grayscale conversion
the significant issue of Lumpy Skin Disease (LSD). Utilizing simplifies the photograph facts through eliminating shade
CNN and SVM, the study develops a model to identify versions, while median filtering successfully gets rid of
different stages of LSD, achieving a remarkable 95.7% unwanted noise.
accuracy. Recommendations include advanced noise removal
techniques for further improvement. B. Image Preprocessing
Image preprocessing accuracy occupies an important place in
5.Rapid Detection of Lumpy Skin Disease Virus: the realization of a deep learning model designed to classify
This study introduces a novel method for on-site detection of nodular skin disease (LSD) in animals This stage assures that
Lumpy Skin Disease Virus (LSDV) using CRISPRCas12a the interpretation gets better information so, constantly from
technology. With high sensitivity and specificity, this assay which it can be checked to improve its detection accuracy.
offers a promising solution for early LSDV detection,
essential for containing outbreaks and minimizing economic The records collected for this study include photographs of
impact. farm animals with and without LSD archived from veterinary
clinics and internet resources Although this section enlarges
the list, it appears as photographs different types will show
III.DETECTION OF LUMPY SKIN DISEASE USING DEEP different definitions of lighting, color intensity, and noise
LEARNING phases. Several preprocessing techniques have been used to
address these inconsistencies and to fine-tune the ability of the
Nodular skin disorder (LSD) poses a significant threat to definition to extract logical functions:
livestock globally, necessitating early and accurate detection
to mitigate economic losses. This section explores the use of Grayscale conversion: To convert images from RGB (Red,
image processing and Convolutional Neural Networks Green, Blue) format to grayscale, you remove the shadow
(CNNs) for LSD detection, focusing on data aggregation and definition which will not work for the LSD type. This
preprocessing techniques. simplifies the facts for the model and reduces computational
requirements.
A. Data Acquisition and Preparation for Lumpy Skin
Disease Detection Mean filtering: Digital images can run the risk of noise for
A important issue of this studies involved accumulating a some time during acquisition, transmission, or storage.
large image dataset to educate and test the Lumpy Skin Median filtering effectively solves this problem by changing
Disease (LSD) detection model. Researchers aimed to acquire the value of each pixel with the value of neighbouring pixels.
a various range of pictures depicting cattle with and with out This method effectively eliminates unwanted noise while
LSD. preserving important image information necessary for LSD
detection.
1.Image Sources:
To acquire this goal, they hired a -pronged approach. Firstly, Contrast enhancement: Subtle differences between healthy
they collaborated with veterinary clinics to gather pics of tissue and LSD and skin may be obscured by changes in
livestock recognized with LSD. This furnished treasured illumination or photorefractive conditions Contrast
statistics representing real-world instances. Second, they techniques such as histograms can be used equalization has
been used to explain these subtle changes in detail. This 8.Identification gadgets for cervical sickness: The results
improves the recognition of key features in the images and strongly endorse that a person has lumpy cancer, if the cancer
enables the interpretation to be very different. is considerate or unsafe.

C. Architecture 9.Input Attributes: All noteworthy attributes, asymmetry, part,

The framework can be comprehensively sorted into following concealment, distance, development, and so forth.
massive ranges: Which have been expelled from the photo at the moment are
furnished as a determination toPart II, that is the classifier
1.Acquisition of photograph: Images are acquired both by part.
means of lens or through secretly deleting them from the
contraction. Whatever the source can be, it is very essential
that the picture of the information is transparent and cautious.
An great photograph is wanted for this.

2.Pre-Processing of picture: In this method, the picture is

standardized by clearing the commotion as it conceals hair
and Bone, as it is able to confuse the evaluation. Similarly, the
photograph given because the statistics might not be of
preferred length as required by means of the figure, so it's far
important that the picture size wanted is acquired.

3.Data storage component to hold records photos for checking Figure 1: System Arcitecture
out and schooling: If controlled studying will occur, as is the
case here, it's miles critical to put together statistics units. The
D. Deep Neural Network
sample database is the pix accumulated at some stage in the
photo procurement technique. The quantity of pix required for Deep Neural Networks (DNNs), additionally referred to as
a given venture is getting large and larger. Algorithms like deep nets, are a sort of artificial neural community structure
convolution neural networks, additionally known as convents used in deep gaining knowledge of. They are inspired by the
or CNNs, can handle huge datasets of photos and even learn shape and feature of the human brain, with a couple of layers
from them. of interconnected nodes that manner information steadily.
Here's a breakdown of key factors of DNNs:
4.Classifier to categorise the kind of cervical sickness: The
classifier used here is the final layer of the device which gives 1.Structure:
the actual probability of each experience. The task includes Unlike less complicated neural networks with only some
predominant elements Image guidance unit and Grouping layers, DNNs have numerous hidden layers stacked among
unit. The item processing device enhances the photograph by the input and output layers. These hidden layers allow the
using eliminating the clatter and noisy bits. The Cervical and community to study complicated, non-linear relationships
the photograph will then be remoted into different segments inside the facts. The more hidden layers a DNN possesses, the
to isolate the lumpy from running the mill after the photo "deeper" it turns into.
functions are evacuated to test whether or now not the lumpy
is infected. 2.Function:
DNNs perform by way of feeding facts thru the input layer,
5.Noise reduction unit: Noise is usually provides in digital where it's miles processed with the aid of the first hidden
photos at some stage in picture acquisition, coding, layer. Each neuron within a layer applies mathematical
transmission, and processing steps. Filtering picture facts is a functions to the statistics it gets from the previous layer, after
general process utilized in nearly every image processing which passes the effects to the following layer. This procedure
machine. Filters are used for this reason. They cast off noise continues thru all hidden layers, ultimately leading to an
from pictures via preserving the info of the same. output at the final layer.

6.Image enhancement unit and segmentation: It incorporates 3.Training:

the affected part to the middle with the aid of improving the DNNs are educated on large datasets of classified examples.
region and dividing the region into distinctive segments in During training, the network adjusts the connections
order to isolate it from the ordinary Scanned Image. (weights) among its neurons based totally at the distinction
between its anticipated output and the actual labeled price.
7.Feature Extraction Components: One of the fantastic traits This iterative process helps the DNN learn how to pick out
in any gathering-targeted troubles is highlighting extraction. patterns and relationships in the facts, permitting it to make
Looks are the cornerstone for both purposes of making plans an increasing number of correct predictions on unseen facts.
and screening. This characteristic carries noteworthy photo
facts a good way to be used to perceive thedisease.
VI. METHODOLOGY
A. Module Specification 2.Feature Extraction
Module Specification involves breaking down the system Feature extraction plays an important position in extracting
into modules to improve structure design. This approach statistics present in given photo. Are using GLCM for texture
reduces complexity and allows testing of modules photo evaluation. GLCM is used to seize spatial dependency
independently. Our model consists of three main modules: among image pixels. GLCM works on grey stage image
pre-processing, identification, and feature extraction and matrix to seize maximum common feature such as contrast,
detection. Each phase represents specific functionalities entropy, energy, homogeneity, correlation, ASM.
provided by the proposed system. In the pre-processing phase,
noise removal using median filtering is conducted. 3. Training:
Model Selection: Choose an in-depth course for the job at
The System design mainly consists of hand. Convolutional neural networks (CNN) are frequently
1.Image Collection used in image classification tasks due to their ability to learn
2.Image Preprocessing hierarchical features.
3.Image Segmentation
4.Feature Extraction Dataset Splitting: Divide the dataset into training, validation,
5.Training and test sets. The training process is used to train the model,
6.Classification the validation process is used to tune hyperparameters and
monitor performance during training, and the testing process
1.Image Collection is used to evaluate the excellent quality of the training.
The dataset used for this project is publicly available on the
internet. It comprises images of various diseases, including Loss Function: Define a loss function that quantifies the
the lumpy disorder dataset. The input to the proposed system difference between the predicted class labels and the ground
is the classification of scanned images and X-ray images. truth labels. For multi-class classification tasks like lumpy
disease classification, commonly used loss functions include
2.Image Preprocessing categorical cross-entropy.
The goal of pre-processing is to enhance image data, reducing Optimization: Choose an optimization algorithm to minimize
distortions and improving image features necessary for the loss function and update the model's parameters iteratively
further processing. It involves three main tasks: during training. Common optimization algorithms include
a)Gray scale conversion Stochastic Gradient Descent (SGD), Adam, and RMSprop.
b)Noise removal
c)Image enhancement Hyperparameter Tuning: Experiment with different
hyperparameters such as learning rate, batch size, and
A) Grayscale conversion: Converting color images into network architecture to improve model performance. This
grayscale simplifies and speeds up processing, as all process often involves conducting multiple training runs with
techniques are applied on grayscale images. In grayscale different parameter settings and selecting the configuration
conversion coloration photograph is converted into grayscale that yields the best results on the validation set.
photo suggests. Grayscale pictures are less difficult and faster
to system than coloured pictures. All photo processing Regularization: Apply regularization techniques such as
approach are carried out on grayscale photograph. dropout and weight decay to prevent overfitting and improve
B) Noise Removal: The goal of noise removal is to stumble the generalization ability of the model.
on and take away undesirable noise from digital picture. The
difficulty is in determining which capabilities of an Training Monitoring: Monitor the training process by tracking
photograph are actual and which are because of noise. Noise metrics such as training loss and validation accuracy over
is a random change in pixel value. We use median filter to epochs. Early stopping can be used to halt training if the
eliminate unwanted noise. Median filtering is a nonlinear performance on the validation set stops improving, thereby
filter that leaves edges unchanged. Central cleaning is done preventing overfitting.
with a special length sliding window. The value of each type
is controlled by values; The mean value is the average of the 6. Classification:
samples in the window that is the output of the filter. A) Inference: Once the model is trained, it can be used to
C) Image Enhancement: The goal of photograph classify new, unseen images. During inference, input images
enhancement is to process an picture to growth visibility of are passed through a training model, resulting in the
function of hobby. Here evaluation enhancement is used to get probability distribution of certain classes. The category with
better best end result. the highest probability is selected as the prediction form for
the input image.
3.Image Segmentation
1.The subsequent step after photograph pre-processing B) Post-processing: In some cases, post-processing
changed into to segment the cervical tumour location from the techniques may be applied to refine the classification results.
encompassing CT Images. A black and white photograph This could involve filtering out false positives, aggregating
became produced with its contrast adjusted to offer higher predictions over multiple image patches, or incorporating
segmentation. domain-specific knowledge to improve accuracy.
C) Performance Evaluation: Evaluate the performance of the positive values and then rejects the negative values. It is
the trained model on the test set using metrics such as the function of low computational cost.
accuracy, precision, recall, and F1-score. These metrics
provide insights into the model's ability to correctly classify
images belonging to different classes of lumpy diseases.

D) Deployment: Once the model has been trained and

evaluated, it can be deployed into production environments
where it can automatically classify lumpy diseases in
realworld scenarios. Deployment considerations include
scalability, latency, and integration with existing systems or
workflows.
Figure 3: Graph of positive ReLU Func
B.CNN Architecture
V. RESULT
1.Convolution layer
Our CNN model achieved promising results in detecting
This layer involves scanning the whole image for patterns and
Lumpy Skin Disease (LSD) in animals. The accuracy,
formulating it in the form of a 3x3 matrix. This convolved
precision, and recall metrics indicate the model's
feature matrix of the image is known as Kernel. Each value in
effectiveness in correctly classifying healthy, mildly infected,
the kernel isknown as weight vector.
and severely infected skin conditions. The visual
representation of the results, through a graph, helps
understand the model's performance across different classes.
Additionally, the grayscale images showcase real-world
examples where the model successfully identified LSD in
animals. These results demonstrate the potential of CNNs for
automated disease detection in livestock, paving the way for
early diagnosis and improved animal health management.
Figure 2: Convolutional Neural Network Layers
B.Pooling layer VI. CONCLUSION
After the convolution comes to the pooling here the image In common, this ponder accomplishes manner better end
matrix is broken down into the sets of 4 rectangular segments result towards vicinity of Uneven skin illness and classify as
which are non-overlapping. There are two types of pooling, Disjoin, Gentle and Ordinary skin. The dedication of this
Max pooling and average pooling. Max pooling gives the recollect carries making plans of Knotty pores and skin
maximum value in the relative matrix region which is taken. malady Picture dataset, improvement of Knotty skin infection
Average pooling gives the average value in relative matrix Picture classification display and Strategy to utilize
region. community facts to recognised frequency of creature Plague
The main advantageof the pooling layer is that it increases malady. The essential undertaking watched on this
computer performance and decreases over- fitting chances. contemplate is the non-life of Knotty pores and skin
contamination Picture records sate for discover and Clamors
for appropriately spotting the locale of intrigued.

VII. REFERENCES
[1]. Abdi Feyisa, (2018), “clinical case studies on major
diseases of veterinary importance in bishoftu town, Ethiopia”,
unpublished Master Thesis, College Of Veterinary Medicine
And Agriculture Department Of Clinical Studies Addis Ababa
university
Figure 3.1: Matrix image of Convolution Layer [2]. Teshome D, Derso S, (2015), "Prevalence of major skin
diseases and their associated risk factors in Ruminants at the
University of Gondar Veterinary Clinic in Northwestern
Ethiopia". -002 J Veterinary Science thiab Technology S: S13-
002. PIB: 10.4172/2157-7579.1000S13
[3]. Ahmed Ali, (2018), “Review on lumpy skin disease and
its economic impacts in Ethiopia”, Journal of Dairy,
Veterinary & Animal Research, Volume 7
[4]. Zeedan GSG, Mahmoud AH, (2019), “Detection of
Figure 3.2: Structure of CNN in matrix format
lumpy skin disease virus in cattle using real-time polymerase
chain reaction and serological diagnostic assays in different
C.Activation layer governorates in Egypt in 2017”, Veterinary World, 12(7):
It the part of the Convolutional Neural Networks where the 1093-1100.
values are Normalized that is, they arefitted in a certain range.
The used convolutional function is ReLU which allows only
[5]. EFSA AHAW Group (EFSA Animal Health and Welfare
Group), (2015), “Scientific opinion on soft skin disease”,
EFSA Journal 2015, 3986, pp. 73.
[6]. Betelihem Tegegne,(2018), “Outbreak investigation of
lumpy skin disease; isolation and molecular characterization
of the virus in south wollo zone, northern Ethiopia”,
unpublished master thesis, college of veterinary medicine and
agriculture, Addis Ababa university
[7]. E.M. El-Nahas and A.S. El-Habbaa, (2011),“Isolation and
Identification of Lumpy Skin Disease Virus from Naturally
Infected Buffaloes at Kaluobia, Egypt”, Global Veterinaria 7
[9]. Gezahegn Alemayehu and Samson Leta, (2015),
“Incidence of lumpy skin disease and associated risk factors
among export-oriented cattle feedlots at Adama District,
Central Ethiopia”, Journal of Veterinary Medicine and
Animal Health, Vol. 7(4), pp. 128-134
[10]. Shubisa Abera, (2017), “molecular characterization of
lumpy skin disease virus isolates from outbreak cases in cattle
from sawena district of bale zone, Oromia, Ethiopia”,
unpublished master thesis, college of veterinary medicine and
agriculture, department of microbiology, immunology
and veterinary public health, Addis Ababa university