Deep Learning-Based Cotton Leaf Disease Detection Using

DenseNet V3

By
Md. Basitur Rahman Bappi, ID:17183103052
Md. Rabbinur Muktar, ID:17183103066
Md. Sakhawat Ullah, ID:17183103070
Sunjida Astania Jerin, ID:17183103037
Umme Sumaya Afrin, ID:17183103053

Submitted in partial fulfillment of the requirements of the degree of Bachelor of Science in

Computer Science and Engineering

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

BANGLADESH UNIVERSITY OF BUSINESS AND TECHNOLOGY

August 2022
 Declaration

We do hereby declare that the research works presented in this thesis entitled “ Deep learning- Based
on Cotton Leaf disease Detection using DenseNet V3” are the results of our own works. We further
declare that the thesis has been compiled and written by us. No part of this thesis has been submitted
elsewhere for the requirements of any degree, award or diploma, or any other purposes except for
publications. The materials that are obtained from other sources are duly acknowledged in this
thesis.

Signature of Authors

Md. Basitur Rahman Bappi Md. Rabbinur Muktar

ID: 17183103052 ID: 17183103066

Md. Sakhawat Ullah Sunjida Astania Jerin

ID: 17183103070 ID: 17183103039

Umme Sumaya Afrin

ID: 17183103053

ii
 Approval
We accept that the research described in this thesis, titled “Help domestic life andcure
diseases by image processing technique for automatic detection” is based on M.M Fazle
Rabbi, Assistant Professor, original work from Bangladesh University of Business and
Technology’s Department of Computer Science and Engineering. We further declare that
no portion of this thesis has ever been submitted to any organization in order to be
considered for a degree, honor, or diploma or for any other reason besides publication
The dissertation's format and content have been approved, and it conforms with all norms
and requirements for the degree of Bachelor of Science (B.Sc. Engg.) in Computer Science
Engineering.

Supervisor Chairman

M. M. Fazle Rabbi Md. Saifur Rahman

Assistant professor Assistant Professor & Chairman (Acting)
Department of Computer Science Department of Computer Science&
Engineering, Bangladesh University & Engineering, Bangladesh University of
Business and Technology of Business and Technology
Dhaka, Bangladesh Dhaka, Bangladesh

iii
 Dedication
We would like to dedicate our research to our loving parents and our respectable supervisor.

iv
 Acknowledgement

We would like to express our heartfelt gratitude to the almighty Allah who offered upon our

family and us kind care throughout this journey until the fulfilment of this research.

Also, We sincerely appreciate and respect our supervisor. Mr. M M Fazle Rabbi, Assistant

Professor, Department of Computer Science and Engineering, Bangladesh University of Business

and Technology (BUBT). Without his direction, this research project would not be possible.We are
grateful to him for his excellent supervision and for putting his utmost effort into developing this

project. We owe him a lot for his assistance, encouragement, and guidance, which has shaped

our mentality as a researcher.

Finally, we are grateful to all our faculty members of the CSE department, BUBT, to

make us compatible to complete this research work with the proper guidance and supports

throughout the last four years.

v
 Abstract

Cotton trees are tropical and subtropical trees that flourish in warm climates. It is a popular,
costly as well as a cash crop. Farmers face difficulties selling their products when their
production is decreased due to diseases harming cotton plants. It is critical to manage any
dangerous illnesses as soon as feasible in order to increase quality and production. This problem
prompted development of novel technologies for detecting and diagnosing cotton plant diseases, as
well as expert systems for disease prevention. The use of computer vision in agriculture has
grown in popularity as a result of its ability to give critical information in real-time. To achieve
our goal, we have built a dataset that included 4 different classes of diseased and leaf images.
We have applied InceptionV3, DenseNet and MobileNetV2 on our dataset. By using DenseNet we
achieve best accuracy of 99.22 percent. Our system with the capability of identifying leaf
diseases of cotton has been built. As a result, we believe that our initiative will help the user by
allowing us to make more items, which will have an impact on our economy.

Keywords: Cotton Leaf Disease, DenseNet, MobileNetV2, InceptionV3, CNN.

vi
List of Tables

4.1 The table provides information about the proposed CNN's validation accuracy like
DenseNet, InceptionV3, and MobileNetV2 approach and CNN architecture........... 28

vii
List of Figures

1.1 Flow of the work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.1 Workflow of the proposed system.................................................................................... 18

3.2 MobileNetV2 Architecture .......................................................................................... 21
3.3 Dense Convolution Network ....................................................................................... 22
3.4 Dense Convolution Network ....................................................................................... 23

4.1 Sample images of the created dataset .............................................................................. 26

5.1 Gantt chart of work execution ......................................................................................... 34

viii
 List of Abbreviation

CLD Cotton leaf disease

CLDD Cotton leaf disease Detection

CNN Convolutional Neural Network

FN False Negative

FP False positive

TN True negative

TP True positive

ix
Contents

Declaration ii

Approval ii
Acknowledgement iii
Abstract iv

List of Tables v
List of Figures vi

List of Abbreviations vi
1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Problem Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.6 Flow of the Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Significance of the Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.8 Research Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.9 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Literature Review 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Problem Analysis ......................................................................................................... 16
2.4 Summary ............................................................................................................................ 16

x
3 Proposed Model 17
3.1 Introduction ....................................................................................................................... 17
3.2 Feasibility Analysis ....................................................................................................... 17
3.3 Requirement Analysis .................................................................................................. 18
3.4 Research Methodology ..................................................................................................... 18
3.4.1 Data Pre-processing .............................................................................................. 18
3.4.2 Convolutional Neural Network ....................................................................... 20
3.4.3 MobileNetV2 .................................................................................................... 21
3.4.4 Dense Convolutional Network ........................................................................ 22
3.4.5 InceptionV3 ...................................................................................................... 23
3.5 Design, Implementation, and Simulation ................................................................... 23
3.6 Summary ............................................................................................................................ 24

4 Implementation, Testing, and Result Analysis 25

4.1 Introduction ....................................................................................................................... 25
4.2 Dataset ............................................................................................................................... 25
4.3 System Setup ............................................................................................................... 27
4.4 Evaluation ..................................................................................................................... 27
4.5 Results and Discussion ................................................................................................ 28
4.6 Summary ............................................................................................................................ 29
5 Standards, Constraints, Milestones 30
5.1 Standards (Sustainability) ............................................................................................ 30
5.2 Impacts (on Society) .................................................................................................... 31
5.3 Ethics .......................................................................................................................31
5.4 Challenges .......................................................................................................................... 31
5.5 Constraints ................................................................................................................... 32
5.6 Timeline and Gantt Chart............................................................................................ 32
5.7 Summary ............................................................................................................................ 35
6 Conclusion 36
6.1 Introduction ....................................................................................................................... 36
6.2 Future Works and Limitations .................................................................................... 36

Appendices 43

xi
 Chapter 1

Introduction

1.1 Introduction

Bangladesh is a developing nation, with agriculture employing over 70 percent of the

population. Farmers have a vast variety of Fruit, Vegetable and Cotton crops from which to
choose. However, cultivating these crops for maximum production and product quality is a
highly technological process. It can be enhanced with technological assistance. Close monitoring is
necessary while managing perennial cotton crops, especially when infections are present that have
the potential to significantly reduce output. the life after harvest as a result. The main source of raw
materials for the country expanding textile industry is cotton, often known as White Gold or King of
Fibers. Cotton is one of the top cash crops in the country. It is an essential agricultural product that
supports millions of farmers in developed and developing nations and provides a living for sixteen
million people. Despite intense competition from synthetic fibers, Bangladesh has a market share of
more than 50% and is the world largest provider of natural fibers. It controls a 70% market share in
the textile industry. Warm-weather crops like cotton (Gossypium spp.) need a steady supply of water,
which can be obtained either naturally through rainfall or artificially through canals from above-
ground and subterranean sources. Despite not being a plant that enjoys being submerged in water,
cotton needs a steady supply of water to support development and a balanced transition between the
vegetative and reproductive phases. About 55% of the world. cotton is grown by irrigation, with the
remainder being watered by rainfall. In

1
Bangladesh, rain feeds 70 percent of the cotton planted land. Agricultural research makes
extensive use of Image RGB feature pixel counting techniques, which have a lot of potential,
particularly in the area of plant protection, which leads to crop management. There are several
applications for image analysis, such as:

I. Detecting damaged leaves, stems.

II. To determine the extent to which a disease has affected an area.

III. Determine the impacted area’s limits.

IV. Find out what color the damaged area is.

V. Determine the size and shape of the Leaf.

VI. To appropriately identify the Object.

This research proposed a sophisticated computational computing system that uses diseased
photos of diverse cotton leaf patches to identify illnesses. The infected portion of the leaf spot was
used to test the proposed MobileNetV2, InceptionV3, DenseNet Algorithm and for train
classification purposes. Images are captured using a mobile digital camera and processed using image
growing. This technology is a system that combines image processing with cutting-edge computing
methods.

1.2 Problem Statement

Analysing Cotton Leaf Disease is one of computer vision’s promising areas. Cotton Leaf disease
analysis becomes more pledging because of sometimes changes in disease behavior analytical power.
Real-time Cotton Leaf Disease analysis and finding disease pattern has remained a challenging and
exciting problem in computer vision. Generating a useful pattern of leaf disease is a complex
problem for deep learning techniques since sometimes different kinds of disease can

2
vary significantly in the way they show their classification. In this work, we design a combined
model to generate significant leaf disease patterns to detect different types of disease, such as
bacterial blight, curl virus, fussarium wilt, and healthy. The data sets available on the internet are
not bringing to perfection because of this classified basic disease of cotton. Therefore, the world has
never seen such cotton leaf disease detection architectures.

1.3 Problem Background

The present Cotton Leaf Disease Recognizer system only classifies these diseases like yellow vein,
Leaf curl, yellow Mosaic, etc. Identifying Leaf disease is a real-world computer vision problem
and yet to solve. Due to the shortcomings of current CLD systems, it is exceedingly difficult to
identify cotton leaf diseases such bacterial blight, curl virus, and fussaraim wilt. Most critically,
there are less datasets available for fussariam, curl virus, and bacterial blight. The main obstacle
to our research is the lack of an appropriate dataset. Working with our own newly developed
dataset is something we look forward to.

1.4 Research Objectives

The objectives of our research work are as follows:

I. Identifying the presence of Cotton Leaf Disease classification.

II. Creating a new dataset named Cotton Leaf Disease (CLD).

III. Reducing computational complexity for Cotton Leaf classification.

IV. Developing a new combined InceptionV3, DenseNet and MobileNetV2 approach to

classify Cotton leaf Disease accurately.

V. Comparing the existing architectures with the proposed one for the Cotton Leaf Disease
classification task.

3
1.5 Motivations

In the last ten years, computer vision researchers have paid a lot of attention to cotton leaf disease
analysis because it resides at the confluence of several important applications, including farmer's
confirming leaf disease analytics. Modern researchers are trying to develop the best approaches to
address these problems. Enormous research work has already done that gives us reasonable
solutions for few CLD problems. However, no specific architecture has developed yet that
identify cotton leaf disease, which significantly impacts our society described in section
5.2. However, to overcome the situations, we built a dataset named Cotton Leaf Disease (CLD)
dataset. Finally, MobileNetV2, DenseNet, InceptionV3 approach was applied to the CLD dataset and
achieve a promising result to identify Leaf disease.

4
1.6 Flow of the Research

The research project is evolving into multiple phases. We first evaluated the potential study
areas before looking at the fundamental concepts underlying Cotton Leaf Disease. After that,
we looked into CLD’s application. We looked at the dearth of existing architectures and inspired
developers to create a brand-new architecture based on cutting-edge deep learning techniques.
The accompanying graphic, Figure 1.1, shows the general steps in the research procedure.

5
 Start

Literature review on
cotton leaf disease
detection

Dataset collects

Identifying new disease

Implement the propose

architecture

Report Writing

End

Figure 1.1. The figure illustrates the flow of the thesis work.

6
1.7 Significance of the Research

We observe We note that the majority of the Cotton Leaf Disease (CLD) picture collections on the
internet were developed for three or four disease classification. However, a dataset broken down into
four classes -bacterial blights, curl virus, fussariam wilt, and health of the leaf was required to analyze
Cotton leaf disease. In order to categorize cotton leaf disease into these four groups, this study offers
a novel dataset. Additionally, we suggested the optimum architecture, MobileNetV2, InceptionV3,
DenseNet. This classification challenge for leaf disease develops into a cutting-edge architecture to
estimate leaf disease and has a big impact on society and the nation.

7
1.8 Research Contribution

The following are the research’s total contributions:

I Identifying present difficulty to analyse Cotton Leaf Disease Detection (CLDD)

problems.

II A There are 1400 photos in a brand-new dataset called Cotton Leaf Disease (CLD)
dataset belongs to 4 different classes.

III It is suggested to classify Cotton Leaf Disease using a new combination of InceptionV3,
DenseNet, and MobileNetV2. This strategy produced compelling results. The combined
strategy is exceptional and is thought to have huge potential. The suggested combined
architecture’s impact is also evaluated in comparison to standard existing architectures.

1.9 Thesis Organization

The organization of the thesis is as follows. The history and literature review on the Cotton Leaf
Disease Detection (CLDD) system are highlighted in Chapter 2.2. The suggested architecture of
the leaf disease classification system and a thorough explanation of the general steps are
presented in Chapter 3. The specifics of the tests and evaluations carried out to assess our
suggested architecture are detailed in Chapter 4. The Standards, Impacts, Ethics, Challenges,
Constraints, Timeline, and Gantt Chart are explained in Chapter 5. Finally, Chapter 6 contains
a summary of our thesis research. Proposed architecture and a detailed walk-through of the
overall procedures. Chapter 4 includes the details of the tests and evaluations performed to
evaluate our proposed architecture. Chapter 5 explains the Standards, Impacts, Ethics,
Challenges, the Constraints, Timeline, and Gantt Chart. Finally, Chapter 6 contains the overall
conclusion of our thesis work.

1.10 Summary
8
This chapter This chapter provides a broad overview of the problem, the context, and the driving
force behind our research activity. It also includes our specific research work objectives. This
chapter also outlines the general procedures we used to conduct our research.

9
 Chapter 2

Literature Review

2.1 Introduction

In the past few years, computer vision researchers have paid a lot of attention to leaf disease
analysis. The majority of existing approaches focus on classifying three or four diseases. Such
as Verticillium Wilt, Leaf blight. However, before this study, no work has introduced on bacterial
blights, curl virus, fusarium wilt, health of the leaf using only InceptionV3, MobileNet, DenseNet
algorithms investigation by analysing leaf diseases. In this paper, we have summarized some of
the research that has been done.

2.2 Literature Review

Over the past few years, researchers studying computer vision have paid close attention to the
analysis of cotton leaf disease detection.

Ch. Usha Kumari, S. Jeevan Prasad, G. Mounika analysis Leaf Disease Detection: K-means
clustering for feature extraction and ANN for classification. The suggested system uses image
processing methods to locate the leaf location. Artificial neural network, K-means clustering.
with its 92.5 percent categorization accuracy on average [1].

Bhumika S.Prajapati, Vipul K.Dabhi, Harshadkumar B.Prajapati analysis A Review of

Cotton Leaf Disease Detection and Classification. In this study, a survey on the identification
and categorization of diseases affecting cotton leaves is presented. They employ methods for
categorization like Support Vector Machine (SVM), Neural Network (NN), and Homogeneous
Pixel Counting approach for Cotton Diseases Detection (HPCCDD). About 80–90% of diseases,
including Alternaria leaf spot and Bacterial blight, primarily affect cotton plant leaves[2].

10
 Supriya S. Patki1, Dr. G. S. Sable2 Detection and Classification of Cotton Leaf Disease
Using Multi SVM. The work is done on cotton leaves in the suggested disease detection system.
They employ a Multi SVM (Multi Support Vector Machine) classifier to apply classification
techniques for identifying disorders. Their accuracy rate was 93.3 percent [3].

1A. Jenifa, 2R. Ramalakshmi, 3V. Ramachandran analysis Deep Convolution Neural
Network for Cotton Leaf Disease Classification for Sustainable Cotton Production. The
detection of cotton leaf disease makes use of the Deep Convolutional Neural Network for the
first time. We evaluate the experimental outcome after training our data, which shows an average
accuracy of 96 percent [4].

P.Revathi, M.Hemalatha Using edge detection techniques for image processing,

classification of cotton leaf spot diseases. They suggested the HPCCDD algorithm, and it had
an accuracy rate of 89.5 percent [5].

H. Ajra, M. K. Nahar, L. Sarkar and M. S. Islam analysis Plant Leaf Disease Detection Using
CNN and Image Processing with Preventive Measures. They are applying image processing
while employing the AlexNet and ResNet-50 models. The trial findings demonstrate the
effectiveness of the suggested strategy, which achieves an overall accuracy of 96.5 percent and
95.3 percent [6].

R.Revathi, M. Hemalatha Analyze a cutting-edge computing technique that has been created to
assist the farmer in making wise decisions regarding various elements of crop development. Using
modern technology methods, cotton leaf spot symptoms were gathered on mobile devices, and
HPCCDD was used to classify the disorders. The accuracy achieved by the suggested method is 98.1
percent. Less data is used for them [7].

11
 P.R. Rothe, R.V. Kshirsagar Gaussian filtering and SVM are used to analyze a pattern to
classify and identify three different cotton leaf diseases. Back propagation neural networks are
used to classify photos of damaged cotton leaves. It is discovered that the categorization
accuracy is 85%. This system processes information very slowly [8].

P.R. Rothe, R.V. Kshirsagar Using a pattern recognition system, classify and identify three
different cotton leaf diseases. By collecting seven invariant moments from three different types
of diseased leaf images, back propagation neural networks are used to classify photos of
diseased cotton leaves. HU's moments are extracted as features for the training of an adaptive
neuro-fuzzy inference system, which uses an active contour model to segment the images. They
employ fewer dataset images [9].

K.Prashaer, R. Talwar, C.Kant The implementation of an expert system for the detection of
crop diseases in the provided set of photographs is the goal of this paper. This study uses an
expert system to identify crop diseases in the provided set of photos. To reduce mistakes through
double-layered modeling, the k-nearest neighbor (kNN) and support vector machine (SVM) are
used for the classification layer that overlaps. With more than 96 percent accuracy, this model
can pinpoint the illness location. The timing parameters need to be improved in this system [10].

A. Jenifa, R.Ramalakshmi, V. Ramachandran The classification of cotton leaf disease using

Multi-Support Vector Machines is suggested. Here, we examine two methods—CNN and Multi-
SVM—to see whether method's accuracy is higher for this system. In comparison to CNN, the
proposed algorithm's classification accuracy is 93.63 percent higher [11].

Sandeep kumar, Arprit Jain, Anand Prakesh Shukla, Satyendr Singh investigation of plant
diseases through comparison utilizing machine learning algorithms. Plant ailments can be
identified and categorized using machine learning replicas. In this study, edge identification and
coordinate histograms are used with SVM and K means algorithms to detect plant diseases. The
accuracy of this approach is 94%. They receive reduced data set usage. [12].
V. A. Gulhane and M. H. Kolekar examine the issue of cotton disease diagnosis

12
leaf utilizing Nearest Neighbourhood Classifier (NNC) and Principle Component Analysis
(PCA) (KNN). It is possible to analyze the statistical information pertaining to the Green (G)
channel of the RGB image after adopting PCA/KNN multi-variable approaches. Green channel
is taken into account for faithful feature collecting since it reflects sickness or elemental deficits
well. PCA/KNN-based classifier's observed classification accuracy is 95% [13].

Bhagya M Patil and Vishwanath Burkpalli study the classification of cotton leaf disease
using a feature-based approach that combines color and texture. Using a modified factorization-
based active contour method, cotton leaf images with a complicated backdrop were recovered.
Color, gray-level co-occurrence matrix, and local binary pattern characteristics were retrieved
from segmented pictures. The feature collection was subjected to machine-learning classifiers
for classification. The proposed model's accuracy is 92.29. They employ fewer data set features
and graphics [14].

R.P.Narmadha, G.Arulvadivu Analysis of Image processing for the detection and

measurement of symptoms of paddy leaf disease the approach created to reduce the amount of
time needed to measure the effects of paddy leaf disease and to eliminate noise automatic human
mistake. Additionally, it improves precision. In segmentation, the image is divided into discrete
sections with each pixel holding characteristics that are comparable. Processed are algorithms
like k-means, FCM, PCA, PSO, and MPSO. The accuracy rate for recognizing the paddy disease
is 94.7 percent. they receive less data set use [15].

Ananda S. Paymode, Shyamsundar P. Magar, Vandana B. Malode employing a

convolution neural network to analyze the detection and classification of tomato leaf
disease. The goal is to accurately identify and categorize illnesses that have already begun
to affect leaves. The goal of our research is to use Deep Learning (DL) approaches to solve
the issue. This analysis requires the best Convolution Neural Network (CNN) method. For
real, collected photos of disease, the CNN-based model's accuracy of 97 percent was the
greatest ever. They used a smaller dataset [16]

Yogita K. Dubey*, Milind M. Mushrif, Sonam Tiple analysis Roughness Based on

Superpixels

13
Measure For Classifying and Detecting Cotton Leaf Diseases. To detect cotton leaf disease and
categorize it according to the type of disease it belongs to, this system offered a computer vision-
based method. With the provided database, proposed methods showed an average classification
accuracy of 94%. less often used dataset [17].

Bhumika S.Prajapati, Vipul K.Dabhi, Harshadkumar B.Prajapati analysis The objective is

to create an automated system that uses image processing to identify the pest or disease affecting
cotton leaf, boll, or blossom. Through this study, we came to the conclusion that changing the
color system from RGB to HSV is effective for background removal. In comparison to other
background removal strategies, we also discovered that the thresholding methodology produces
good results. By masking the green pixels in the background-removed image and performing
otsu thresholding on the resulting masked image to produce a binary image, we performed color
segmentation. This helps to accurately extract disease-related characteristics. In terms of
accuracy, we discovered that SVM performs well for classifying disorders. less often used
dataset [18].

Yiqing Guo, Feng Zhao, Yanbo Huang, Matthew A. Lee, Krishna N. Reddy, Reginald
S. Fletcher, Steven J.Thomson, Jianxi Huang The soybean and cotton leaves were used in a
study effort to invert foliar biochemical parameters in order to detect crop damage from
glyphosate. Through non-destructive hyperspectral reflectance measurements of soybean and
cotton leaves, PROSPECT, a physically based leaf radiation transport model, is reversed. The
process of gathering data, testing it, and finally coming to the best option was one of the most
difficult aspects of this study. less often used dataset [19].

Mr. Ketan D. Bodhe, Mr. Himanshu V. Taiwade, Mr. Virendra P Yadav, Mr. Nikesh
V. Aote They proposed using a straightforward web site that farmers could simply visit in order
to identify and diagnose cotton leaf disease. This web portal is accessible from anywhere at any
time using a mobile device and an internet connection. In this instance, template matching
techniques are being used to find cotton leaf disease. User selects an image first, and the
algorithm then compares it to an image of a disease. The system then displays the leaf condition.
less often used dataset [20].

14
 P. Sivakumar, N. Sri Ram Mohan, P. Kavya, P. Vinay Sai Teja transfer learning models were
built using ResNet50, VGG19, InceptionV3, and ResNet152V2 CNN architectures. According
to experimental findings, ResNet50, VGG19, InceptionV3, and ResNet152V2 are providing
75.76%, 87.64%, 96.46%, and 98.36% of the results, respectively. ResNet152V2 with
parameters 60,380,648 outperformed the other three models in terms of accuracy. ResNet152V2
is a transfer learning approach that delivers better accuracy and is therefore a highly beneficial
notion for disease identification in plants [21].

Shubham Bavaskar, Vaibhav Ghodake, Gayatri Deshmukh, Pranav Chillawar, Atul Kathole
employing deep learning-based image classification to analyze crop disease detection systems. By
scanning cotton plant leaves, the technology can identify the three cotton diseases Bacterial Blight,
Curl Virus, and Fusarium Wilt. With the ResNet152 V2 architecture, the system's best accuracy was
99.12% on the training dataset and 98.6% on the testing dataset. they employ fewer datasets [22].

Bhagya M Patil, Vishwanath Burkpalli Using a modified factorization-based active contour

method, cotton leaf images with a complicated backdrop were recovered. Color, gray-level co-
occurrence matrix, and local binary pattern characteristics were retrieved from segmented pictures.
The proposed model has a 92.29 percent accuracy rate. They employ fewer dataset images [23].

Nikhil Shah, Sarika Jain The main goal of the study is to identify the various diseases that affect
cotton by using an artificial neural network tool, which processes images using an image pre-
processing method and highlights the main area of the affected leaf in the image based on color
changes to identify the disease type. Using the RGB and HSV components from the image, the
technique was able to identify cotton leaf with or without flaws. they utilize less datasetet images
[24].

Yan-Cheng Zhang, Han-ping Mao, Bo Hu Utilizing computer vision to recognize and treat
cotton illness The design and functionality of the classifier are impacted by feature selection, a
crucial pattern recognition issue in agriculture. To choose features of cotton disease leaves,
fuzzy curves and surfaces are offered as part of the fuzzy feature selection approach image.
They employ fewer photos from the dataset[25].

15
 Youssef Es-saady, Ismail El Massi, Mostafa El Y assa, Driss Mammass, 2 Abdeslam
Benazoun presents a machine learning technique for automatically identifying illnesses on plant
leaves from photographs. The serial combination technique of two SVM classifiers is the
foundation of the suggested system. Two SVM classifiers are serially combined to form the
suggested technique. This system has an accuracy rate of 87.80 percent. They employ less
dataset photos and features[26].

Aditya Parikh, Mehul S. Raval, Chandrasinh Parmar, Sanjay Chaudhary two cascaded
classifiers are being analyzed. Leaf is separated from the background using a first classifier
using local statistical features. Another classifier is then trained to identify diseases and
determine their stages using hue and brightness from the HSV color space. The created
algorithm is versatile and can be used to treat any illness. This system has an accuracy rate of
82.5 percent. They employ less dataset photos and features[27].

Kavita Bhosle, Vijaya Musande Red Edge Point identifies the portion of the observation that
is green. Red Edge Point is present in the vegetation as determined by remote sensing techniques.
Different crops can be identified using the trial data. Remote sensing data can be used to locate
unsound crops. Refraction and reflection data from a spectrum radiometer are similar to that of
remote sensing data. If we can locate Red Edge Point, this would be achievable. This method
can be used to determine whether a plant is dry. Finding plant stress in mulberry, cotton, and
sugarcane plants is the current focus of this paper. The results are estimated utilizing peak
derivative, linear interpolation, and linear extrapolation methods. They employ fewer dataset
images [28].

Adhao Asmita sarangdhar, V. R. Pawar They are demonstrating a technique for monitoring
soil quality as well as for detecting and eradicating infections on cotton leaves. The article
suggests a regression approach based on support vector machines for identifying and
categorizing five illnesses of cotton leaves. This system's proposed accuracy rate is 83.26
percent [29].

G. Harshitha S. Kumar S. Rani A. Jain Due to the disease's invasion, cotton production
decreased last year. In most cases, these pests or insects that cause plant diseases

16
If not controlled quickly, cotton plant leaf diseases can significantly lower yield; they must be
detected early and accurately. In this study, a deep learning model that was trained on both clean
and infected photos is suggested. On the Cotton Disease Dataset Kaggle, the suggested solution
performs better than existing state-of-the-art algorithms, with a classification accuracy of 97.13
percent. They employ fewer dataset images [30].

However, all these CLDD techniques only analysed the three or four kind of diseases but
did not give any solution to new diseases detection. In this paper, we have discussed how the
MobileNetV2 , InceptionV3, DenseNet combined approach solves this issue.

2.3 Problem Analysis

Cotton leaf disease detection Over the past few years, analysis has drawn a lot of attention from
computer vision experts. these CLDD techniques only analysed the three or four kind of diseases
such as fusarium wilt, Bacterial blights, verticillium wilt, leaf blight but did not give any solution
to new diseases like curl virus which we have successfully done in this project with better
accuracy. In this thesis work, we have discussed how the Mobile NetV2, InceptionV3, DenseNet
approach solves this issue.

2.4 Summary

This chapter investigated and reviewed the latest techniques of Cotton leaf disease detection systems,
including the drawbacks. The thesis’s target is to eliminate the imperfections as muchas possible
and introduced a new approach to investigate cotton leaf disease.

17
 Chapter 3

Proposed Model

3.1 Introduction

The feasibility study of cotton leaf disease using cotton leaf image processing is supported in this
area, along with the demands made in this structure. This chapter concludes by illuminating the
model's overall architecture, which is provided through a thorough explanation.

3.2 Feasibility Analysis

This study project took eleven months to develop and involved five researchers and one supervisor.
Technical assistance, including hardware and software, was needed for the thesis work. The
researchers also carried out the dataset development and evaluation processes that were necessary
for the research project. The enormous amount of data required for the project is gathered while
taking its legal viability into account. The institution and supervisor did not need to provide any
financial assistance for the thesis study.

18
3.3 Requirement Analysis

To carry out the suggested architecture's general needs,

I. Device for High-performance computing system.

II. Picture input mechanism.

III. library of free software for scientific computations.

IV. Open-source software tools for putting the deep learning paradigm into practice.

3.4 Research Methodology

In This section elaborates on the suggested architecture's technique. There are four components
within this section. The sub-sections are arranged sequentially from the model's input to output phase
with thorough justifications. Additionally, Figure 3.1 displays the architecture's overall workflow.

Input
Disease Detection
Processing the Input to classify
dataset CNN image

Figure 3.1. The suggested system's workflow is shown in the figure (from left to right).

3.4.1 Data Pre-processing

Data normalization and data splitting were the two stages of the data pre-processing. The
following describes these two methods.

19
 Data Normalization:
Image normalization is a crucial pre-processing technique. It lessens the difference in inner-class
features and is regarded as intensity offsets. The local area has preset intensity offsets. Gaussian
normalization and the standard deviation are therefore helpful when normalizing. Equation calculates
the final image after normalizing (3.1) [31].
ξ (π, θ) − µ (π, θ)
ψ (π, θ) = .......................................................... (3.1)
6σ (π, θ)

Where µ is a local mean and σ is a local standard deviation [32].

1 Σ
α Σ
α

µ (π, θ) = ε (K + µ, n + θ) ......................................... (3.2)

M 2 k=−α n=−α
‚
. 1 Σ α Σα
µ (π, θ) = , [ε (K + µ, n + θ) − µ (π, θ)]2 ................. (3.3)
M 2 k=−α n=−α

Data
splitting:

In order to increase the network resistant mutation in input samples, we perform a variety of
transformations on the training and testing samples during assessment. The CPU is used for this
image mutation technique while the GPU is used for network training and evaluation. In order
to gain greater accuracy, deep learning architectures always demand a large number of input
samples. Our CLD dataset, which includes 1708 photos for 4 classes, is insufficient to assess a
deep learning architecture.

When data is divided into two or more subgroups, this is known as data splitting. A two-part split
often involves testing or evaluating the data in one part and training the model in the other.
Data splitting is a crucial component in data science, especially when building models from data.
This method aids in ensuring the accuracy of data model construction and the processes that use data
models, such as machine learning.

20
3.4.2 Convolutional Neural Network

We define the data pre-processing technique that can be used with any shape and quality of input
because photos of cotton leaf disease come in a variety of sizes and qualities. In this design, CNN
builds two dense layers with ReLu activation functions and six convolutional layers with dropout for
training. The operations of convolution and completely linked layers are described in equations (3.4)
and (3.5). Additionally, we used regularization for each weight matrix, which reduces the volume of
the weights at each layer to a few fixed values. The regularization procedure is described in Equation
(3.5).

(l −1)
m
Yi (l)
=B (l)
Σ
i
K(l)Y (l−1)
..................................... (3.4)
+
i i,j j
j=1

where the output Y (l)

i
of layer l consists of the m(l) feature of size m(l) × m(l). The ith
3 1 2
(l) (l) (l)
feature map denoted Y and B is a big bias matrix and K is the filter of size.
i i i,j

d (x) = Activation W T x + b .............................................. (3.5)

where, W = [W1, W2, ....Wn]T represents the weight vector of the dense layer, and b
represents the bias value of the dense layer.

ReLU = max(0, x).................................................. (3.6)

if the ReLU function gets any non-positive value, it returns zero but for any positive input
of x, it returns that input value.

x, with prob.p
Dropout(x, p) =
(3.7)
x, with prob.1 − p

where x be the output of a particular neuron in the network and p the dropout possibility.
21
3.4.3 MobileNetV2

MobileNetV2 is a convolutional neural network architecture that seeks to perform well on

mobile devices. It is based on an inverted residual structure where the residual connections are
between the bottleneck layers. The intermediate expansion layer uses lightweight depthwise
convolutions to filter features as a source of non-linearity. As a whole, the architecture of
MobileNetV2 contains the initial fully convolution layer with 32 filters, followed by 19 residual
bottleneck layers.[33]

Figure 3.2. MobileNetV2 Architecture

22
3.4.4 Dense Convolutional Network

A DenseNet is a type of convolutional neural network that utilizes dense connections between
layers, through Dense Blocks, where we connect all layers (with matching feature-map sizes)
directly with each other. To preserve the feed-forward nature, each layer obtains additional
inputs from all preceding layers and passes on its own feature-maps to all subsequent layers.[34]

Figure 3.3. Dense Convolution Network

23
3.4.5 InceptionV3

InceptionV3 is a convolutional neural network architecture from the Inception family that makes
several improvements including using Label Smoothing, factorized 7 x 7 convolutions, and the
use of an auxiliary classifier to propagate label information lower down the network (along with
the use of batch normalization for layers in the side head) [35]

Figure 3.4. Dense Convolution Network

3.5 Design, Implementation, and Simulation

Figure 3.1 depicts the proposed architecture's total workflow. Python is used to carry out every
stage of the prototype described above. Keras is used to create convolutional neural network
models like DenseNet, InceptionV3, and MobileNetV2. Numpy [36] is also utilized for further
calculation, implementation, and support. There were no changes or selections made to the
dataset used to evaluate the architecture; it was just directly added.

24
3.6 Summary

The suggested cotton leaf disease classification method's architecture is described in this section.
Convolutional neural networks, such as DenseNet, InceptionV3 and MobileNetV2, are incorporated
in the overall design.

25
 Chapter 4

Implementation, Testing, and Result Analysis

4.1 Introduction

This section explains the architecture of the proposed cotton leaf disease classification method. The
overall architecture uses the combined convolutional neural network like DenseNet, Inception V3,
MobileNetV2.

4.2 Dataset

We note that the majority of the cotton leaf disease (CLD) image datasets available online were
developed for the classification of three or four different cotton leaf diseases. However, a dataset
with four classes—bacterial blights, curl virus, fussarium wilt, and health of leaf—is required in
order to analyze leaf disease. As a result, we created our cotton leaf disease (CLD) dataset by
using web gathering techniques to collect photographs of cotton leaf disease from the internet.
As previously mentioned, we add four classes to our dataset: leaf health, fussarium wilt, curl
virus, and bacterial blights. We have gathered a huge collection of pictures from the same category.
Finally, for the four classes, we placed 1708 photos. 60 percent (1024) of the photos were used for
training during assessments, while the remaining 40 percent (684) were used for testing. Four classes'
photos are spread evenly, and there are 450 samples in each class. A

26
Figure 4.1. The sample images of Blacterial blights, curl virus, fussarium wilt, healthy from
the dataset that we used for this study, seen from top to right. The online gathering method was used
to collect these photographs from the internet.

Figure 4.1 displays a few representative photos from the CLD dataset.

27
4.3 System Setup

For pre-processing data, experimenting, and model evaluations, Python is the programming language
of choice. TensorFlow and Keras are used to implement the suggested architecture. NumPy is
additionally utilized to do mathematical operations on the architecture.

4.4 Evaluation

On our CLD dataset, we undertake experimental research to assess the offered solutions for the
classification difficulties related to cotton leaf diseases. The influence of the combined
MobileNetV2, InceptionV3, and DenseNet technique to classifying cotton leaf diseases was
investigated in this study.

To determine how effective an algorithm or strategy is, comparative and transferable

performance measures are necessary. Adopting training and testing sets, which might cause an
inconsistency in model performance, is the main challenge for evaluating any approach. The
confusion matrix, which includes true-positive (TP), true-negative (TN), false-positive (FP), and
false-negative (FN) values [37], is the foundation for the majority of performance indicators.
Depending on how the performance review is conducted, the importance of these components
may change.

The number of accurate guesses the model predicts from all of the model's estimations serves
as a measure of an identification system's accuracy. The precision is evaluated using,

TP + TN
Accuracy ................................................ (4.1)
= TP + TN + FP +
FN

Precision indicates how many of the positive classes that the model properly predicted are
actually positive. We divide the total number of successfully classified positive cases by the total
number of anticipated positive examples to get the precision value. The formula can be written as,

28
4.5 Results and Discussion

We implement our CNN-like Mobile Net, Inception V3, and Dense Net architecture using the
CLD dataset. Table 4.1 displays the CNN's prediction accuracy, precision, and recall like Mobile
Net, Inception V3 and Dense Net approach for classifying leaf diseases of cotton. This table
illustrates how the suggested architecture performs and contrasts the suggested architecture's
anticipated outcome with CNN like MobileNetV2, InceptionV3 and DenseNet architecture. Our
investigation found that using Three kind of architectures like MobileNetV2, Dense net,
InceptionV3 architectures does classify the leaf disease properly. MoblieNetV2 architecture
gives only 0.9734 accuracy and InceptionV3 architecture gives only 0.9743 accuracy and
DenseNet architecture gives only 0.9922 accuracy. It should be noted that the DenseNet technique
performs better than the MobileNetV2, DenseNet, and InceptionV3 designs in terms of performance
evaluation. Therefore, DenseNet accuracy is more successful in producing superior results. Each
outcome in this study is presented as the mean over four runs. 100 training epochs are set as the
maximum for the suggested model.

Model Accuracy
DenseNet 99.22
MobileNetV2 97.34
Inception V3 97.43

Table 4.1. The table displays the predicted CNN's accuracy in terms of validation. like
DenseNet, InceptionV3, and MobileNetV2 approach and CNN architecture.

29
 This study examined at the CNN's importance in great detail like DenseNet, InceptionV3
and MobileNetV2 combine approach to detection cotton leaf disease and founds satisfactory
performance.

4.6 Summary

According to the evaluation analysis, this architecture delivers the best results for detecting
illnesses in cotton leaves.

30
 Chapter 5

Standards, Constraints, Milestones

The Standards, Impacts, Ethics, and Challenges of the thesis study are illustrated in this section.
The limitations and options are then illustrated. Finally, the proposed work’s schedules, tasks,
and milestones are described.

5.1 Standards (Sustainability)

We make sure that our thesis work is durable and long-lasting. Analysis of the cotton leaf
disease is a current, well-liked research topic. Detecting leaf disease is a CLD issue that can
be beneficial for the nation and society. Additionally, the CNN implementation we employed,
along with DenseNet, InceptionV3, and MobileNetV2, represents the state-of-the-art in edge
deep learning. We may say that this thesis work will be sustainable because the materials we
employed will be available for longer stretches of time.

31
5.2 Impacts (on Society)

There are numerous diseases that exist in our culture. With a view to classification, various
diseases are divided into.

Consequently, one of the most crucial components of a plant disease detection is diagnosis.
Proper disease diagnosis is therefore vital. In our current world, due to climate change, various
diseases are occurring in plants, which farmers cannot determine with their eyes. Also, most
farmers around the world are poorly educated, making diagnosis difficult for them. For which
they can’t give the correct treatment for plants disease. Here in our proposed system, it can
classify the disease and give the right treatment statement for the disease with accuracy of
99.22 percent, which will be as a blessing for the farmers.

5.3 Ethics

The (CLDD) Cotton Leaf Disease Detection system has a vast application area, recline on the
dataset applied to train the model. The installation of the system enhances a social, national or
global impact on plant saving. The collection of the dataset must be complete under the code of
moral principles and ethics.

5.4 Challenges

As farmers don’t have much idea on technology, if they don’t give proper resolution images for
disease detection, for image quality it will be difficult for the system to detect the disease. Sonot
giving a proper image, the processing of disease detection can be difficult or challenging for the
system.

32
5.5 Constraints

Different constraints such as design constraints, component constraints, and budget constraints are
presented in this section.

Based on the image dataset, a proposed general structure is constructed. A powerful

processor is required to efficiently handle a huge number of photos for our modal. However, a
GPU is not necessary. This component is used to train our model:

I. Minimal processor: Intel Core i3 (7th gen)

II. Minimal memory: 4GB (DDR4, 2400bus)

III. Image Input: Image Input Device

However, because the price of the product component is not constant, the budget may change
depending on the market condition.

5.6 Timeline and Gantt Chart

Due to the fact that we have three semesters to complete our work, our timeframe for the thesis is
separated into three halves. We followed our supervisor's instructions for how to complete our
assignment. We filed a proposal and examined the thesis work's associated work during the first
semester. In addition, we studied and planned with the existing systems before building the prototype
of the proposed systems. In the second semester, we developed a dataset and partially applied the
model. Last but not least, in the third semester, we put the overall design into practice, tested it using
the dataset that was introduced, and reported the entire workflow.

The procedure of carrying out the work is shown in the following Gantt Chart (Figure 5.1).

33
the thesis projects. The thesis work is finished in three semesters, with a semester lasting four
months and a total of 12 months.

34
 1st Semester

Weeks 1 2 3 4 5 6 7 8 9 10 11 12

Planning and Study

Topic Selection

Review Related

work

Analysing Existing Systems

Built Prototype

Evolution
2nd Semester

13 14 15 16 17 18 19 20 21 22 23 24
Weeks

Model

Diagram Model &

System DesignDesign

Submission Model

Analysis

Partial Implementation

Evolution 3rd Semester

25 26 27 28 29 30 31 32 33 34 35 36

Weeks

Complete Implementation & Testing

Check Issue &

Resolved Model

Finalization Result

Evaluation Report

Writing

Presentation & Final Evaluation

Figure 5.1. Gantt chart of the work execution process.

5.7 Summary

However, this chapter briefly explains the standards, impacts, ethics, challenges of the thesis work.
Also, the constraints, alternatives, schedules, tasks, and milestones of the proposed work are
demonstrated.

35
 Chapter 6

Conclusion

6.1 Introduction

The dataset gathered from Kaggle is used in this study to test and assess the Cotton Leaf Disease
classification algorithm. To precisely train and test our system, we used InceptionV3,
mobileNetV2, and DenseNet techniques. We note that the categorization of Cotton Leaf Disease
performs better using the DenseNet technique. We believe that the architecture described in this
work is the first to categorize curl virus using the DenseNet method. Additionally, the
performance of the suggested architecture is contrasted with that of other architectures.

6.2 Future Works and Limitations

Our thesis is the first research work on curl virus detection using DenseNet from still images to
the best of our knowledge. Cotton Leaf Disease detection from using deep learning architectures can
be a significant research field. Besides, we have used DenseNet to develop the architecture, but the
latest another architecture can be tested for the task. We strongly believe that “Cotton Leaf Disease
Detection “Classification Using DenseNet, MobileNetV2, InceptionV3 The research project
"Approach" will open the door for considerable study on Cotton Leaf Disease Detection and improve
the knowledge and applicability of future work.

36
References

[1] Ch. Usha Kumari, S. Jeevan Prasad, and G. Mounika. Leaf disease detection: Feature

extraction with k-means clustering and classification with ann. In 2019 3rd International
Conference on Computing Methodologies and Communication (ICCMC), pages 1095–1098,

2019.

[2] Bhumika S. Prajapati, Vipul K. Dabhi, and Harshadkumar B. Prajapati. A survey on

detection and classification of cotton leaf diseases. In 2016 International Conference on

Electrical, Electronics, and Optimization Techniques (ICEEOT), pages 2499–2506, 2016.

[3] Supriya Patki and Ganesh Sable. Ijarcce cotton leaf disease detection classification using multi

svm. 10 2007.

[4] A. Jenifa, R. Ramalakshmi, and V. Ramachandran. Cotton leaf disease classification

using deep convolution neural network for sustainable cotton production. In 2019 IEEE

International Conference on Clean Energy and Energy Efficient Electronics Circuit for

Sustainable Development (INCCES), pages 1–3, 2019.

[5] P. Revathi and M. Hemalatha. Classification of cotton leaf spot diseases using image

processing edge detection techniques. In 2012 International Conference on Emerging

Trends in Science, Engineering and Technology (INCOSET), pages 169–173, 2012.

[6] Husnul Ajra, Mst. Khairun Nahar, Lipika Sarkar, and Md. Shohidul Islam. Disease detection

of plant leaf using image processing and cnn with preventive measures. In 2020

37
 Emerging Technology in Computing, Communication and Electronics (ETCCE), pages

1–6, 2020.

[7] P. Revathi and M. Hemalatha. Classification of cotton leaf spot diseases using image

processing edge detection techniques. In 2012 International Conference on Emerging

Trends in Science, Engineering and Technology (INCOSET), pages 169–173, 2012.

[8] Viraj A. Gulhane and Maheshkumar H. Kolekar. Diagnosis of diseases on cotton leaves

using principal component analysis classifier. In 2014 Annual IEEE India Conference

(INDICON), pages 1–5, 2014.

[9] P. R. Rothe and R. V. Kshirsagar. Cotton leaf disease identification using pattern

recognition techniques. In 2015 International Conference on Pervasive Computing (ICPC),

pages 1–6, 2015.

[10] A. Jenifa, R. Ramalakshmi, and V. Ramachandran. Classification of cotton leaf disease

using multi-support vector machine. In 2019 IEEE International Conference on Intelligent

Techniques in Control, Optimization and Signal Processing (INCOS), pages 1–4, 2019.

[11] Kapil Prashar, Rajneesh Talwar, and Chander Kant. Cnn based on overlapping pooling

method and multi-layered learning with svm knn for american cotton leaf disease recogni-
tion. In 2019 International Conference on Automation, Computational and Technology

Management (ICACTM), pages 330–333, 2019.

[12] Anand Prakesh Shukla Sandeep kumar, Arprit Jain. A comparative analysis of machine

learning algorithms for detection of organic and nonorganic cotton diseases. In Hindawi

Mathematical Problems in Engineering 2021, 2021.

[13] Viraj A. Gulhane and Maheshkumar H. Kolekar. Diagnosis of diseases on cotton leaves

using principal component analysis classifier. In 2014 Annual IEEE India Conference
(INDICON), pages 1–5, 2014.

38
[14] Bhagya M Patil and Vishwanath Burkpalli. Cotton leaf disease classification by combining

color and texture feature-based approach. In 2022 6th International Conference on

Intelligent Computing and Control Systems (ICICCS), pages 1783–1789, 2022.

[15] R. P. Narmadha and G. Arulvadivu. Detection and measurement of paddy leaf dis- ease

symptoms using image processing. In 2017 International Conference on Computer

Communication and Informatics (ICCCI), pages 1–4, 2017.

[16] Ananda S. Paymode, Shyamsundar P. Magar, and Vandana B. Malode. Tomato leaf disease

detection and classification using convolution neural network. In 2021 International

Conference on Emerging Smart Computing and Informatics (ESCI), pages 564–570, 2021.

[17] Yogita K. Dubey, Milind M. Mushrif, and Sonam Tiple. Superpixel based roughness

measure for cotton leaf diseases detection and classification. In 2018 4th International

Conference on Recent Advances in Information Technology (RAIT), pages 1–5, 2018.

[18] Bhumika S. Prajapati, Vipul K. Dabhi, and Harshadkumar B. Prajapati. A survey on

detection and classification of cotton leaf diseases. In 2016 International Conference on

Electrical, Electronics, and Optimization Techniques (ICEEOT), pages 2499–2506, 2016.

[19] Yiqing Guo, Feng Zhao, Yanbo Huang, Matthew A. Lee, Krishna N. Reddy, Reginald S.

Fletcher, Steven J. Thomson, and Jianxi Huang. Early detection of crop injury from

glyphosate by foliar biochemical parameter inversion through leaf reflectance measurement.

In 2013 Second International Conference on Agro-Geoinformatics (Agro-Geoinformatics),

pages 116–120, 2013.

[20] Ketan D. Bodhe, Himanshu V. Taiwade, Virendra P. Yadav, and Nikesh V. Aote. Implemen-

tation of prototype for detection diagnosis of cotton leaf diseases using rule based system for

farmers. In 2018 3rd International Conference on Communication and Electronics

Systems (ICCES), pages 165–169, 2018.

[21] P. Sivakumar, N. Sri Ram Mohan, P. Kavya, and P. Vinay Sai Teja. Leaf disease

39
 identification: Enhanced cotton leaf disease identification using deep cnn models. In 2021

IEEE International Conference on Intelligent Systems, Smart and Green Technologies

(ICISSGT), pages 22–26, 2021.

[22] Shubham Bavaskar, Vaibhav Ghodake, Gayatri Deshmukh, Pranav Chillawar, and Atul

Kathole. Image classification using deep learning algorithms for cotton crop disease

detection. In 2022 IEEE International Conference on Distributed Computing and Electrical

Circuits and Electronics (ICDCECE), pages 1–8, 2022.

[23] Bhagya M Patil and Vishwanath Burkpalli. Cotton leaf disease classification by combining
color and texture feature-based approach. In 2022 6th International Conference on

Intelligent Computing and Control Systems (ICICCS), pages 1783–1789, 2022.

[24] Nikhil Shah and Sarika Jain. Detection of disease in cotton leaf using artificial neural
network. In 2019 Amity International Conference on Artificial Intelligence (AICAI), pages

473–476, 2019.

[25] Yan-Cheng Zhang, Han-Ping Mao, Bo Hu, and Ming-Xi Li. Features selection of cotton

disease leaves image based on fuzzy feature selection techniques. In 2007 International

Conference on Wavelet Analysis and Pattern Recognition, volume 1, pages 124–129, 2007.

[26] Youssef Es-saady, Ismail El Massi, Mostafa El Yassa, Driss Mammass, and Abdeslam

Benazoun. Automatic recognition of plant leaves diseases based on serial combination

of two svm classifiers. In 2016 International Conference on Electrical and Information

Technologies (ICEIT), pages 561–566, 2016.

[27] Aditya Parikh, Mehul S. Raval, Chandrasinh Parmar, and Sanjay Chaudhary. Disease

detection and severity estimation in cotton plant from unconstrained images. In 2016 IEEE

International Conference on Data Science and Advanced Analytics (DSAA), pages594–

601, 2016.

[28] Kavita Bhosle and Vijaya Musande. Red edge point detection for mulberry leaf. In 2017 1st

40
 International Conference on Intelligent Systems and Information Management (ICISIM),

pages 119–122, 2017.

[29] Adhao Asmita Sarangdhar and V. R. Pawar. Machine learning regression technique for

cotton leaf disease detection and controlling using iot. In 2017 International conference of

Electronics, Communication and Aerospace Technology (ICECA), volume 2, pages 449–454,

2017.

[30] G. Harshitha, S. Kumar, S. Rani, and A. Jain. Cotton disease detection based on deep

learning techniques. In 4th Smart Cities Symposium (SCS 2021), volume 2021, pages
496–501, 2021.

[31] Neha Jain, Shishir Kumar, Amit Kumar, Pourya Shamsolmoali, and Masoumeh Zareapoor.

Hybrid deep neural networks for face emotion recognition. Pattern Recognition Letters,

115:101–106, 2018. Multimodal Fusion for Pattern Recognition.

[32] S.M. Zhang and X.B. Li. Spatial distribution characteristics and regional influencing factors

of characteristic towns in china. Resource Development Market, 8, 2018.

[33] Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen.

Mobilenetv2: Inverted residuals and linear bottlenecks. 2018.

[34] Gao Huang, Zhuang Liu, Laurens van der Maaten, and Kilian Q. Weinberger. Densely

connected convolutional networks.

[35] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna.

Rethinking the inception architecture for computer vision, 2015.

[36] Stefan Van Der Walt, S Chris Colbert, and Gael Varoquaux. The numpy array: a structure

for efficient numerical computation. Computing in Science & Engineering, 13(2):22–30,

2011.

41
[37] Kamran Kowsari, Mojtaba Heidarysafa, Tolu Odukoya, Philip Potter, Laura Barnes, and

Donald Brown. Gender detection on social networks using ensemble deep learning.

Advances in Intelligent Systems and Computing, 04 2020.

42
Appendices

43