PLANT DISEASE CLASSIFICATION

USING DEEP LEARNING

Dissertation submitted in fulfilment of the requirements for the Degree of

MASTER OF TECHNOLOGY
in
COMPUTER SCIENCE AND ENGINEERING

By
HIMANSHI RAJPUT
11602541

Supervisor
MS. SURUCHI TALWANI

School of Computer Science and Engineering

Lovely Professional University
Phagwara, Punjab (India)
January – May 2021

@ Copyright LOVELY PROFESSIONAL UNIVERSITY, Punjab (INDIA)

ALL RIGHTS RESERVED

1
2
 ABSTRACT

Plant is the common source of food for living. But there are various plants that gets
diseased at its early stage. The spread of this disease at large scale can harm the
economy condition of the country and can lead to the scarcity of specific kind of food
at an instance. Plants being severely diseased can be detected through the leaves of
the plants and can be treated at the earlier stage of growth of the plant. Plants mainly
suffers from disease like Late Blight, Early Blight, Mold, Common Rust, Mold and
many more. These diseases can be observed in the leaves part of the plant which are
mainly caused by the viruses, fungi, and bacteria. The recent advancement in
technologies like Machine Learning, Deep learning, Image Processing Techniques
and Optimization techniques has helped in the early detection and classification of the
plant disease. There are various Deep Learning Model that has been developed by
researchers like Alex Net, Google Net, Deep CNN, Dense Net, ResNet50 etc., models
have been developed. These models have been developed to improve the accuracy of
classification of the model. This paper focuses on considering the large dataset of
images of multiple plants. Various disease of plant has been considered. Then, a Deep
Neural Network has been constructed using various layers like Convo2D, Max-
Pooling Layer, Flatten Layer, Dense Layer, Fully Connected Layer, Drop Out Layer
etc. Hyper-Parameter tuning has been done to improve the accuracy of the model. The
accuracy of the model obtained is 98.13 percentage.

3
 DECLARATION STATEMENT

I hereby declare that the research work reported in the dissertation/dissertation

proposal entitled "PLANT DISEASE CLASSIFICATION USING DEEP
LEARNING” in partial fulfilment of the requirement for the award of Degree for
Master of Technology in Computer Science and Engineering at Lovely Professional
University, Phagwara, Punjab is an authentic work carried out under supervision of
my research supervisor Mr./Mrs. Research Guide’s Name. I have not submitted this
work elsewhere for any degree or diploma.

I understand that the work presented herewith is in direct compliance with

Lovely Professional University’s Policy on plagiarism, intellectual property rights,
and highest standards of moral and ethical conduct. Therefore, to the best of my
knowledge, the content of this dissertation represents authentic and honest research
effort conducted, in its entirety, by me. I am fully responsible for the contents of my
dissertation work.

Signature of Candidate

Himanshi Rajput

R.No. 11602541

4
 SUPERVISOR’S CERTIFICATE

This is to certify that the work reported in the M.Tech Dissertation/dissertation

proposal entitled “PLANT DISEASE CLASSIFICATION USING DEEP
LEARNING”, submitted by Himanshi Rajput at Lovely Professional University,
Phagwara, India is a bonafide record of his / her original work carried out under my
supervision. This work has not been submitted elsewhere for any other degree.

Signature of Supervisor

Ms. Suruchi Talwani

Date:
Counter Signed by:

1) Concerned HOD:
HoD’s Signature: ________________
HoD Name: ____________________
Date: ___________________

2) Neutral Examiners:

External Examiner

Signature: _______________

Name: __________________

Affiliation: ______________

Date: ___________________

Internal Examiner

Signature: _______________

Name: __________________

Date: ___________________

5
 ACKNOWLEDGEMENT

I would like to convey my most heartfelt and sincere gratitude to my guide MS.
SURUCHI TALWANI of Lovely Professional University, for her valuable guidance
and advice. I have taken efforts in this project. However, it would not have been
possible without the kind support and help of many individuals and organizations. I
would like to extend my sincere thanks to all of them. I am highly indebted to Lovely
Professional University for their guidance and constant supervision as well as for
providing necessary information regarding the project & also for their support in
completing the project.
I would like to express my gratitude towards the Head of Department and other
faculty members of Computer Science and Engineering Department, LPU for their
intellectual support through this entire work as well as my parents for their kind co-
operation and encouragement which help me in completion of this project. I would
like to express my special gratitude and thanks to the research scholars for giving me
such attention and time. Finally, I indebted to all my colleagues, dear & nears and all
whosoever has contributed in this dissertation work directly or indirectly.

HIMANSHI RAJPUT
M.Tech(CSE)

6
 TABLE OF CONTENTS

CONTENTS PAGE NO.

Inner first page – Same as cover i

PAC form ii

Abstract iii

Declaration by the Scholar iv

Supervisor’s Certificate v

Acknowledgement vi

Table of Contents vii

List of Acronyms / Abbreviations (If any) viii

List of Symbols (If any) ix

List of Figures x

List of Tables xi

CHAPTER1: INTRODUCTION 12

1.1 TYPES OF DISEASE 13

1.2 DATASET 19

1.2.1. TRAIN DATASET 22

1.2.2. VALID DATASET 23

CHAPTER2: REVIEW OF LITERATURE 26

CHAPTER3: PRESENT WORK 29

3.1 PROBLEM FORMULATION 29

3.3 OBJECTIVES OF THE STUDY 30

7
 TABLE OF CONTENTS

CONTENTS PAGE NO.

3.1 RESEARCH METHADOLOGY 31

CHPTER4: RESULTS AND DISCUSSION 40

4.1 EXPERIMENTAL RESULTS 30

CHAPTER5: SUMMARY AND CONCLUSION 44

REFERENCES 45

8
 LIST OF ACRONYMS/ABBREVIATIONS

Abbreviation FULL FORM

DL Deep Learning
ML Machine Learning
DNN Deep Neural Network
DCNN Deep convolutional Neural Network

9
 LIST OF TABLES

TABLE NO. TABLE DESCRIPTION PAGE NO.

Table 1.1 Train dataset 22

Table 1.2 Valid Dataset 23
Table 1.3 DataSet Details 25
Table 1.4 Hyper Parameter of CNN 37

10
 LIST OF FIGURES/IMAGES

FIGURE NO. FIGURE DESCRIPTION PAGE NO.

Figure1.1 Apple Scab Disease 14

Figure1.2 Apple Black Rot Disease 14
Figure1.3 Cherry Powdery Mildew 15
Figure1.4 Corn Common rust 15
Figure1.5 Pepper Bell Bacterial Spot 16
Figure1.6 Tomato Bacterial Spot 16
Figure1.7 Tomato Early Blight 17
Figure1.8 Tomato Late Blight 17
Figure1.9 Tomato Leaf Mold 18
Figure1.10 Tomato Spider mites (Two spotted spider mites) 18
Figure1.11 Image obtained using vertical flipping 20
Figure1.12 Image obtained using 30-degree rotation 20
Figure1.13 Image obtained using 90- degree rotation 21
Figure1.14 Product between linear and non-linear function 31
Figure1.15 Graph of ReLu Activation Function 33
Figure1.16 Max-Pooling Layer 34
Figure1.17 Flattern Layer in Deep CNN 37
Figure1.18 Visualization of train dataset 41
Figure1.19 Visualization of valid dataset 41
Figure1.20 Training and validation accuracy 42
Figure1.21 Training and validation loss 42

11
 CHAPTER 1

INTRODUCTION

Plants and crops are considered as the most important source of food for animals and
human being. It is also considered as the source for economic development.
Degradation in the quality of plants and crops will result into economic loss. The
degradation in quality can be caused because of the disease in plants. This disease is
caused because of using lowest quality of seeds or because of the change in
temperature and the weather condition. The research in the plant disease detection and
classification has developed a great interest There are different kinds of plants that
gets diseased because of various conditions like weather, temperature condition.

Identifying plant disease has been the important area of research in precision
agriculture. Precision agriculture deals with the observing of crops and measuring the
severity of the factors affecting the crops and taking appropriate actions. Detection of
the plant disease is very tough for the farmers without the help of any expert advice.
They experience lot of problem in finding the kind of disease and taking appropriate
measures to prevent the loss of the crops. Also, the availability of the expert at every
place during need is impossible. Due to this, the farmers face a huge loss in the
production of the crops as they are not able to take right decision at the right time.
Performing manual detection of the disease in crops can be either time consuming or
the decision made by experts may be less accurate. So, the developed technologies
have developed various methods and algorithms for the automatic’s detection and
identification of the type of disease without the need of expert.

Advancement in technology has helped to detect and identify the disease of plant and
resulted in more precise agriculture practice. This advancement has resulted into in
technologies like image processing, artificial intelligence, Graphical processing unit,
deep learning (branch of artificial intelligence) has helped to classify the disease in
plants more accurately. Various Techniques like Machine Learning, Deep Learning,
Fuzzy Neural Network, Clustering Algorithm, Expert System, KNN, SVM, Decision
Tree, Random Forest, Back Propagation method etc., and other techniques has already
been used in identification and classification of disease. Deep Learning is actually an

12
automatic detection technique that has created an advancement in precision
agriculture.

The plant which are diseased need to be treated at very early stage. Some plants get
diseased at very early stage and need to be treated with proper amount to pesticides.
The more severe the disease, it requires more pesticides to treat. It is easier to detect
the disease in plants using the images of the plants and the deep learning method, as
Deep Learning methods and the number of layers in it helps to easily extract the
features from the images

1.1 Types of Plant Disease

There are various kind of images of different plants and its various kinds of
disease has been considered. The different kinds of plants, whose images has been
considered are:

1. Apple

2. Blueberry

3. Cherry

4. Corn(maize)

5. Grape

6. Orange Haunglongbing

7. Peach

8. Pepper

9. Potato

10. Raspberry

11. Soybean

12. Squash_ Powdery mildew

13. Strawberry

13
14. Tomato.

Here are some of the images showing various kinds of diseases that could exist in
the leaves of the plant.

1.1.1. Apple Scab Disease

Figure 1.1 Apple scab Disease

1.1.2. Apple Black Rot

Figure 1.2 Apple Black Rot Disease

14
1.1.3. Cherry Powdery Mildew

Figure 1.3 Cherry Powdery Mildew Disease

1.1.4. Corn Common Rust

Figure 1.4 Cherry Powdery Mildew Disease

15
1.1.5. Pepper Bell Bacterial Spot

Figure 1.5 Pepper Bell Bacterial Spot Disease

1.1.6. Tomato Bacterial Spot

Figure 1.6 Tomato Bacterial Spot Disease

16
1.1.7. Tomato Early Blight

Figure 1.7 Tomato Early Blight Disease

1.1.8. Tomato Late Blight

Figure 1.8 Tomato Late Blight Disease

17
1.1.9. Tomato Leaf Mold

Figure 1.9 Tomato Leaf Mold Disease

1.1.10. Tomato Spider mites (Two-spotted spider mites)

Figure 1.10 Tomato Spider mites (Two spotted spider mites) Disease

18
 There is disease that can be found in some kind of plant only, whereas the other
kind of disease can be found in the other kind of plant. For example, the disease
considered in case of Apple plant are Apple_scab, Apple_Blackrot,
Cedar_apple_rust. Also, the images of the healthy leaves of the plants have also
been considered. The disease considered in Apple plant may not be found in
Tomato Leaves. So, for tomato plant the disease like bacterial_spot, Early_blight,
Late_blight, Leaf_mold, Septoria_LeafSpot, Spidermites (two spotted spider
mites), Target_spot, Tomato_mosaic_virus and Tomato_Yellow_Leaf_Curl_virus
has been considered.

1.2 Dataset

The dataset of the plant disease images has been considered from Kaggle. The
data itself contains pre-processed images. The dataset contains the images of
diseased leaf as well as healthy leaf. There are total of 14 kinds of plants which
has been considered in the complete image dataset of plants. There 14 kinds of
plants have it’s own kind of diseases.

The images of leaves of plants like Apple, Blueberry, Cherry, Corn (maize),
Grape, Orange_Haunglongbing, Peach, Pepper, Potato, Raspberry, Soybean,
Squash, Strawberry, Tomato has been considered in the dataset.

The dataset contains three kinds of data: train dataset, valid dataset, test dataset.

1.2.1. Train Dataset

Training dataset will be used to train the proposed model. Training dataset also
contains images which has been obtained by data augmentation methods. Some of
images contained in the training dataset has already been applied to data
augmentation method. Data augmentation method help to regularize the dataset by
slightly modifying the dataset by using some of the data augmentation techniques.
These methods help us to increase the dataset size and helps to create the synthetic
dataset from the existing data set. Data Augmentation method helps to reduce the
overfitting problem while training the deep neural network model. Here, we are
oversampling the dataset for image data analysis. There is various method

19
a. Flipping: The images in the dataset has been flipped horizontally or
vertically. Horizontal flipping refers to obtaining the mirror image of the
diseased image along horizontal axis. Similarly, Vertical Flipping refers to
the obtaining the mirror image of the image dataset along the vertical axis.

Figure 1.11 Image obtained using Vertical flipping

b. Rotation: The images can be obtained by rotating the images at certain

angle. The image rotation can be done by rotating the images at different
angles like by 30-degree, 90 degree or 270 degree.

Figure 1.2 Image obtained using 30 Degree Rotation

20
 Figure 3 Image obtained using 90 Degree Rotation

c. Scaling: The image can be scaled inward and outward. The image can be
either zoomed in or zoomed out. It is either scaled to obtain the image
bigger than the original image or smaller than the original image.

d. Translation: In this technique, the image is shifted from its original

position. It is either shifted horizontally along x-axis, or vertically along y-
axis or to some co-ordinate (x,y).

e. Cropping: In this method the set of images is selected, cropped, and then
resized to the original image. This technique also helps to increase the
image data size.

Here, are the completer details regarding the type of disease and the number of
disease of particular kind of disease.

Table 1.1 Train Dataset

21
S. PLANT PLANT DISEASE Actual
No. count

1. Apple Apple_scab 2016

Apple__Black_rot 1987

Cedar_apple_rust 1760

Healthy 2008

2. Blueberry Blueberry_healthy 1816

3. Cherry Cherry_(including_sour)_healthy 1826

Cherry_(including_sour)_Powdery_mildew 1683

4. Corn_(maize) Corn_(maize)__Cercospora_leaf_spot_Gray_leaf_spot 1642

Corn_(maize)__Common_rust_ 1907

Corn_(maize)__Healthy 1859

Corn_(maize)__Northern_Leaf_Blight 1908

5. Grape Grape__Black_rot 1888

Grape__Esca(Black measeles) 1920

Grape__Healthy 1692

Grape__Leaf_blight_(Isariopsis_leaf_Spot) 1722

6. Orange__Haunglongbing Orange__Haunglongbing_(Citrus_greening) 2010

7. Peach Peach__Bacterial_spot 1838

Peach__healthy 1728

8. Pepper Pepper,_bell__Bacterial_spot 1913

Pepper,_bell__healthy 1988

9. Potato Potato__Early_blight 1939

Potato__healthy 1824

Potato__Late_blight 1939

10. Raspberry Raspberry__healthy 1781

11. Soybean Soybean__healthy 2022

12. Squash__Powdery_mildew 1736

13. Strawberry Strawberry__healthy 1824

22
 Strawberry__Leaf_scroch 1774

14. Tomato Tomato__Bacterial_spot 1702

Tomato__Early_blight 1920

Tomato__healthy 1926

Tomato_Late_blight 1851

Tomato_Leaf_Mold 1882

Tomato__Septoria_leaf_spot 1745

Tomato__Spider_mites Two-spotted_spider_mite 1741

Tomato__Target_Spot 1827

Tomato__ Tomato_mosaic_virus 1790

Tomato__ Tomato_Yellow_Leaf_Curl_Virus 1961

1.2.2. Valid Dataset

Table 1.2 Valid Dataset

S. PLANT PLANT DISEASE Actual

No. count

1. Apple Apple_scab 504

Apple__Black_rot 497

Cedar_apple_rust 440

healthy 502

2. Blueberry Blueberry_healthy 454

3. Cherry Cherry_(including_sour)_healthy 456

Cherry_(including_sour)_Powdery_mildew 421

4. Corn_(maize) Corn_(maize)__Cercospora_leaf_spot_Gray_leaf_spot 410

23
 Corn_(maize)__Common_rust_ 477

Corn_(maize)__Healthy 465

Corn_(maize)__Northern_Leaf_Blight 477

5. Grape Grape__Black_rot 472

Grape__Esca(Black measeles) 480

Grape__Healthy 423

Grape__Leaf_blight_(Isariopsis_leaf_Spot) 430

6. Orange__Haunglongbing Orange__Haunglongbing_(Citrus_greening) 503

7. Peach Peach__Bacterial_spot 459

Peach__healthy 432

8. Pepper Pepper,_bell__Bacterial_spot 478

Pepper,_bell__healthy 497

9. Potato Potato__Early_blight 485

Potato__healthy 456

Potato__Late_blight 485

10. Raspberry Raspberry__healthy 445

11. Soybean Soybean__healthy 505

12. Squash__Powdery_mildew 434

13. Strawberry Strawberry__healthy 456

Strawberry__Leaf_scroch 444

14. Tomato Tomato__Bacterial_spot 425

Tomato__Early_blight 480

Tomato__healthy 481

Tomato_Late_blight 463

Tomato_Leaf_Mold 470

Tomato__Septoria_leaf_spot 436

24
 Tomato__Spider_mites Two-spotted_spider_mite 435

Tomato__Target_Spot 457

Tomato__ Tomato_mosaic_virus 448

Tomato__ Tomato_Yellow_Leaf_Curl_Virus 490

The Complete Dataset contains:

Table 1.3 Dataset Details

DATASET NO. OF IMAGES

TRAIN SET 70295

VALID SET 17572

TOTAL 87867

25
 CHAPTER 2

LITERATURE REVIEW

Susmita Mishra et al. [1] [2020] has trained Deep CNN Model using diseases images
of corn plant and that has been compiled and tuned model on NCS. The real time
images have been used as test images to test the training model. Various Deep Neural
Network layers has been used to create the model. Various layers like Convolutional
Layer, Max-Pooling, drop out Layer, Flatten Layer and Dense Layers has been used
to construct the model.

Anjna et al. [2] [2020] has proposed a hybrid system for the detection and
classification of plant (capsicum) disease like Bacterial Canker, Bacterial Spot,
Anthracnose, Cercospora leafspot, Gray Spot and Powdery mildew.

Agarwal et al. [3] [2020] has proposed a deep neural network for the detection of the
tomato leaf disease detection. The author has constructed a convolutional neural
network using various convolutional layer, max pooling layer, drop-out rate and
hyper-tuned various other parameters like activation function, learning rate,
momentum, number of epochs, Batch Size etc. The author has obtained 91.2%
accuracy by considering the dataset of 9 different types of Tomato disease and
considering one healthy class of Tomato Plant. The author has then compared the
proposed model with other training models like Mobilenet, VGG16, Inception V3
models.

Ji et al. [4] [2020] has proposed a multi convolutional neural network which is
UnitedModel Based and has considered only grape leaf disease. The author has
considered various grape disease like black rot, isariopsis leaf spot and esca diseases.
The author has constructed a UnitedModel using Inception V3 and ResNet50, then
these both the training model is merged and some of the layer like drop-out layer,
Dense Layer, Softmax layer has been used. The proposed model is then evaluated
using Precision, F1 and recall score. Confusion matrix analysis has been done for the
proposed model. The result is then compared with other existing models like
VGGNet, GoogLeNet, ResNet model. The proposed model has obtained 99.17
percent training accuracy and 98.57 testing accuracy.

26
Rahman et al. [5] [2020] has worked on identification and recognition of rice disease.
The author has used convolutional neural network and obtained the desired frequency
of 93.3 percent. The dataset has been collected from Bangladesh Rice Research
Institute (BRRI). The dataset contains almost 1426 images of the diseased rice leaves
and contains 8 and 1 other kind of rice leaf disease. The convolution neural network
contains 5 convolutional layers and contains activation layer, Batch normalization
layer, Max-poling layers in between the convolutional layers. The author has also
used Dense layers and drop out layer as well.

Picon et al. [6] [2019] has proposed a conditional convolutional neural network for
various crop plant disease. The trained model is then integrated with the mobile
system and where the classification of the real time image from the crop field can be
done over cell phone. The model has been trained over some specific crop diseases
like Winter wheat, corn, Rape seed, Winter Barley, and common rice. The Proposed
convolutional neural network is constructed using transfer learning from ResNet 50
and multiple other layers like Average Pooling layer, fully connected layer, and
Sigmoid function has been used in the activation layer. The proposed method focuses
on removing 71% classification error while training the deep convolution neural
network.

Jayme et al. [13] [2019] has focused more on the lesion and spot portion of the leaf,
rather than considering the whole leaf. This paper has focused on the identification of
multiple disease affecting the same leaf. Image processing technique has been used
along with the Deep learning technique has also been applied on the leaves of plants
like Common bean, cassava, citrus, coconut tree, corn, coffee etc.

Kamal et al. [16] [2019] have used the image dataset of 82,161 images and applies
depth wise separable convolutional Neural Network and obtains the accuracy of
98.65% using reduced MobileNet Deep Learning Model. It also contains 29 times
lesser parameter than VGG and 6 times lesser parameter than MobileNet.

Muhammad et al. [17] [2018] has focused on recognition and classification of fruit
disease. Deep Features and correlation coefficient have been used. The author has
proposed a technique to extract the feature and then classify. Two pre-trained models
VGG16 and caffe AlexNet has been used to extract the feature of selected disease.

27
The dataset has been taken from PlantVillage. The proposed model gives an accuracy
of 98.60%.

Marko et al. [19] [2018] has used Neural Network and Deep Learning technique on
the dataset of 79,265 images from plant village. This dataset contains the diseased
image as well images of healthy leaf. Approximately all the important disease has
been included in the study resulting into 93.67% accuracy. He has used Data
Augmentation method to increase the dataset of images by rotating, scaling, flipping
the images. The author has applied Geometric Transformation (Cropping, Fliiping,
Resizing, Rotation) as well as Intensity transformation (Adjusting brightness, Noise
Removal, Zooming, Segmentation). General Adversial Network (GAN) has been used
for generating syntactic data from existing datasets. Different models like AlexNet,
VGG 19, DenseNet, ResNet has been used for trained model performance.

28
 CHAPTER 3

PRESENT WORK

The training model has been constructed using the layers. There are already existing
models like GoogLenet, AlexNet, ResNet, VGG16, etc. which has been constructed
using multiple of these following layers. Deep Learning can be used to easily classify
the images with higher accuracy. Different model has already been proposed by
researchers which shows good accuracy. Some of these models are further combined
to achieve higher accuracy. This research work intends to use these layers and hyper
tune some of the parameters to increase the accuracy of the model.

3.1 PROBLEM FORMULATION

A lot of research has been done in the field of plant disease detection using Image
processing technique, Convolutional Neural Network, Deep Learning, Using Back
Propagation Algorithm, SVM algorithm, Clustering technique, Particle Swarm
Optimization technique to increase the efficiency and accuracy of the detection of
disease.
This research works aims:
• to extract the features of the images of the plant using deep convolutional
neural network model from the images, in order to detect and classify the
disease at the earlier stage of the growth.
• Deep Convolutional Neural Network will be used to train the model and
classify the type of disease with higher accuracy than before.
• Various other research has been done already to detect the plant disease and
classify the diseases as well. There was some research which worked on a
smaller number of plant disease or have specifically worked on particular
Plant. But this research will focus on achieving the better accuracy of the
algorithm after deep learning algorithm for the feature extraction
• the aim is to achieve more accurate result with increasing the data size of plant
leaves.

29
 • Research has been done for the detection of the plant disease, but it has not
been implemented at large scale in the agriculture field due to which still a lot
of losses has been faced by the farmers. Accuracy of 91% has already been
achieved using feature extraction and neural network whereas 94.67 % has
been achieved using Artificial Neural Network and K-mean algorithm.
• This research intends to increase the accuracy. Before the researchers have
applied the algorithm and have discovered better result for some disease but
for some other disease the accuracy obtained was 50%only. So, this research
will intend to fill that gap for all kind of disease and increase accuracy from
the base paper.

3.2 OBJECTIVES OF THE STUDY

The study is done to increase the efficiency and output with the proposed model.
Detecting and classifying accurate disease will help the farmers to take accurate step
to stop the spread of disease at the stage of early growth of the plant.

The objective of the research work will be as follows:

• To consider large dataset of plant disease images to train the proposed model.
• To use the data set that contains the images of various kinds of crops and deals
with various kinds of disease of particular kind of plants.
• The DCNN model will be using multiple layers like convolutional layer,
flatten layer, drop out layer, flatten layer, fully connected layer, dense layer
etc.
• To include more layers, already proposed model will be used. Some of the
proposed models are AlexNet, GoogLeNet, VGG 16, ResNet and many others
are available.

The proposed model will be used to classify the disease of plant and the comparison
of the proposed model will be done with other classification techniques to show its
accuracy.

30
So, the main objective is to propose the model with higher accuracy and result in less
classification time. This will help the farmers to get the accurate output in lesser time.

3.3 RESEARCH METHODOLOGY

1. Convolutional Layer

There are multiple dimensional convolutional layers. There exist

convolutional 1D layer, convolutional 2D layer, and convolution 3D layer as
well. In case of constructing a deep convolutional neural network, the
convolutional layer used is 2D. Convolution 2D layer is the spatial
convolution over images. In this layer the kernel is convolved with the inputs
of the images to produce the tensor of the output.

This layer of the convolutional neural network performs the product of two
matrices, namely set of learnable parameters (which is also known as kernel)
and the. This layer helps to extract the features from the input images. The
convolutional 2-D mapping is done for every 2-D images. Each image is
mapped into a convolutional window and the corresponding convolution
image is obtained after the convolution operation (*) between the linear
operator and non-linear activation function. The output obtained after
convolutional layer is passed into next layer.

Figure 1.14 Product between linear and non-linear function

31
 Here, Wi+1 represents the linear operation

And xi represents the input.

There is various non- linear activation function that can be used. It helps us to
place the non-linearity after the convolutional layer. These activation function
introduces the non-linearity to the activation map. Some of the non-linear
operations that can be used are:

a. Sigmoid

b. Hyperbolic Tanh

c. ReLu

d. Softmax

The activation function that has been used is ReLu.

ReLu Activation function: ReLu activation function has been used as compared
to sigmoid and hyperbolic tanh, because it is more reliable and accelerates the
convergence faster (by 6 times). The model also takes less time to train itself.

It can be computed using the following function. The activation is threshold at

zero. ReLu activation function is fragile during training.

It helps us to deal with the vanishing gradient problem which may occur in the
other activation functions like sigmoid or hyperbolic tanh activation function.
ReLu activation function outputs zero for all negative inputs.

32
 Figure 1.15 Graph of Relu Function

2. Max-Pooling Layer

This layer has been used to aggregate the statistics of the feature extracted
from the convolutional layer. It considers the maximum value from each patch
of the feature map.

It helps us to reduce the total number parameters to train the model. The total
number of parameters obtained through the layers are 5,809,062. By using the
pooling layer, the total number of parameters that has been reduced to
5,804,070 are trainable. There are 4,992 non trainable parameters. Therefore,
pooling layer helps to reduce the number of parameters to learn. The use of
pooling layer is to reduce the dimension of the feature map. Pooling layer
helps us to make the deep neural network model robust to the variation in the
position of the features in the input image.

Among pooling layers, there are various of its types. There is max pooling
layer, Average Pooling layer, Global Pooling. Among these three types, Max-
Pooling layer has been used to select the maximum number of features from
the feature map covered by the filter. It contains the most prominent feature
from the feature map which has been obtained from the previous layers.

33
 Figure 1.16 Max-Pooling Layer

Advantages of Max-Pooling layer:

• It helps in the over-fitting by providing an abstract form of the

representation.

• It reduces computational cost. It reduces the number of parameters to

learn and train the model with a smaller number of parameters.

• It provides invariance to the representation of the feature map.

3. Batch Normalization Layer

Batch Normalization Layer helps in building an efficient model. It is added

between the layers of deep neural network. This layer takes the input from the
previous layer, normalizes it, and passes the output to the next layer. Batch
Normalization layer is the regularization technique.

Advantage of Batch Normalization Layer:

• It is added to sequential model to standardize the input and output.

34
 • This layer helps us to normalize the inputs, regularize and can be
adopted to converge better.

• It is used to maintain the distribution of data.

• Batch Normalization layer helps to train the model faster.

• Batch Normalization layer is also used to increase the accuracy of the

model and provide better learning.

• It helps to reduce the internal covariate shift.

• It also helps to reduce the effect of overfitting and functions well with
generalized data.

4. Drop Out Layer

• In this some neurons are omitted in the network after specific iteration,
in order to obtain a different neural network for the mini batch of the
training data.

• This layer is used to prevent the overfitting in the deep neural network
model.

• It is the regularization techniques.

• It helps to reduce the capacity of the network by thinning the neural

network by training.

5. Activation Layer

• The activation layer is used to trandform the weighted sum of the input
images into the output node in alyer of network.

• It helps to perform complex functional mapping between unput and

output variables

• It introduces nonlinear operation between the layers.

35
 • In activation layer, activation function is passed, and the activation
function used is ReLu.

• Multiple activation layer has been used in between the other deep
neural network layer.

• ReLu activation function has been used in the activation layer because
it help to remove the vanishing gradient problem.

• Vanishing Gradient is a problem that arises when we increase the

number of stacked layers(hidden layers) in the network. This makes
the gradient descent very slow and decrease exponentially to zero
because of updating the weights at each layer while performing back
propagation from last layer to the input layer. It is very time
consuming.

• The problem of vanishing gradient can’t be resolved by using

traditional machine learning technique. So the ReLU activation
function has been used in the activation layer.

6. Flatten Layer

• Flatten Layer is used to retain the channel dimension and collapse the
spatial dimension of polled features.

• This layer adds an extra dimension. Flattening can also reduce the
number of classes by using softmax activation.

• It helps to flatten the output layer of the convolution neural network

and reates the single long vector of features.

• Figure 3.3 Max-Pooling Layer

36
 Figure 1.17 Flatten Layer in DCNN

7. Dense Layer

• Dense Layer is the fully connected layer.

• All the features, that has been extracted are used for the classification are fully
connected at the end.

• These are fully connected layers that perform linear operations.

Here, is the sequential model that has been constructed using the following number of
parameter and the hyper tuned parameter been used.

Figure 1.4 Hyper Parameter Tuning in CNN

Hyper Parameter Description

No. of Convolutional 2D Layers 5

Number of Batch Normalization Layer 7

Number of activation Layer 7

37
Number of Max-Pooling Layer 5

Number of Dropout Layer 7

Number of Flatten layer 1

Number of dense Layer 3

Learning rate 0.0005

Epoch 25

Drop Out Rate 0.25

Activation Function ReLu

Here, is the completer sequential model constructed using various layer and the
possible trainable parameter and non trainable parameter at each layer.

model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 48, 48, 64) 1792
_________________________________________________________________
batch_normalization (BatchNo (None, 48, 48, 64) 256
_________________________________________________________________
activation (Activation) (None, 48, 48, 64) 0
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 24, 24, 64) 0
_________________________________________________________________
dropout (Dropout) (None, 24, 24, 64) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 24, 24, 128) 204928
_________________________________________________________________
batch_normalization_1 (Batch (None, 24, 24, 128) 512
_________________________________________________________________
activation_1 (Activation) (None, 24, 24, 128) 0
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 12, 12, 128) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 12, 12, 128) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 12, 12, 512) 590336
_________________________________________________________________
batch_normalization_2 (Batch (None, 12, 12, 512) 2048
_________________________________________________________________
activation_2 (Activation) (None, 12, 12, 512) 0
_________________________________________________________________

38
max_pooling2d_2 (MaxPooling2 (None, 6, 6, 512) 0
_________________________________________________________________
dropout_2 (Dropout) (None, 6, 6, 512) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 6, 6, 512) 2359808
_________________________________________________________________
batch_normalization_3 (Batch (None, 6, 6, 512) 2048
_________________________________________________________________
activation_3 (Activation) (None, 6, 6, 512) 0
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 3, 3, 512) 0
_________________________________________________________________
dropout_3 (Dropout) (None, 3, 3, 512) 0
_________________________________________________________________
conv2d_4 (Conv2D) (None, 3, 3, 512) 2359808
_________________________________________________________________
batch_normalization_4 (Batch (None, 3, 3, 512) 2048
_________________________________________________________________
activation_4 (Activation) (None, 3, 3, 512) 0
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 1, 1, 512) 0
_________________________________________________________________
dropout_4 (Dropout) (None, 1, 1, 512) 0
_________________________________________________________________
flatten (Flatten) (None, 512) 0
_________________________________________________________________
dense (Dense) (None, 256) 131328
_________________________________________________________________
batch_normalization_5 (Batch (None, 256) 1024
_________________________________________________________________
activation_5 (Activation) (None, 256) 0
_________________________________________________________________
dropout_5 (Dropout) (None, 256) 0
_________________________________________________________________
dense_1 (Dense) (None, 512) 131584
_________________________________________________________________
batch_normalization_6 (Batch (None, 512) 2048
_________________________________________________________________
activation_6 (Activation) (None, 512) 0
_________________________________________________________________
dropout_6 (Dropout) (None, 512) 0
_________________________________________________________________
dense_2 (Dense) (None, 38) 19494
=================================================================
Total params: 5,809,062
Trainable params: 5,804,070
Non-trainable params: 4,992

39
 CHAPTER 4

RESULTS AND DISCUSSION

4.1 EXPERIMENTAL RESULT

Figure 1.18 Visualization of the Training Dataset

Figure 1.18 shows the various kind of disease of plants. Total classes obtained were
38. The Deep Neural Network was used to train the Deep Convolutional Neural
Network by using these number of images of the diseased as well healthy plant
40
leaves. The Dataset of the images has benn obtained by using data augmentatioon
techniques. The flipping of images, horizontally, vertically and the rotation of the
images has been done using 30 degree, 90 degree, 270 degree and 360 degree as well.

Figure 1.19 Visualization of the Valid Dataset

Figure 1.19 show the dataset visualization of the valid dataset which has been used for
the validation of the dataset while training the deep neural network model by using
training the dataset. It also contains total of 38 classes. There are total of 14 kinds of
plants. The Valid dataset contains diseased as well as the healthy leaf images.

41
 Figure 1.20 Training and validation accuracy

The Deep Neural Network Model has been trained using the above dataset and
contain 38 classes. The model has been trained by using 25 number of epochs ans
shows 98.13 percent of accuracy at the end of training the model. This model is then
used to test and classify the kind of disease in the test data set. The graph shows that
while training the model, the accurcy has been increasing significantly. More
accuracy can be obtained by using more number of epochs while training the model
using the dataset.

42
 Figure 1.21 Training and validation loss

The Deep Neural Network Model has been trained using the above dataset and
contain 38 classes. The model has been trained by using 25 number of epochs ans
shows 98.13 percent of accuracy at the end of training the model. The loss percentage
obtained at the end of the training of model was 5.66 percentage. This model is then
used to test and classify the kind of disease in the test data set. The graph shows that
while training the model, the accurcy has been increasing significantly and the
Training and validation loss has been decreasing significantly. More accuracy can be
obtained by using more number of epochs while training the model using the dataset.
Lesse the training and validation loss, more accurate the classification result wull be.

43
 CHAPTER 5

SUMMARY AND CONCLUSION

Recognizing and classification plant disease is very important in precision agriculture.

Disease in plants can result into vast loss. Farmers and experts should know the
disease and its severity on time. The disease should be detected at the early stage of
the growth of the plant and should be treated with enough pesticides.

Data set that was considered was large and contained more than 80,000 images.
The images were obtained using Data Augmentation method. The dataset had good
review on the Kaggle website. Considering large dataset helped in increasing the
accuracy with real-time dataset and avoid the overfitting problem. Proposed Model
was constructed using various number of layers. The layers that were used to
construct the deep convolutional neural network were Convolution2D layer, Max
poolin2D layer, Activation Layer, Drop Out layer, Dense Layer, Flatten Layer. The
Layer helped to reduce the Vanishing gradient problem. These layers also helped to
avoid the Internal Covariate Shift. Hyper tuning the parameters helped to obtain the
higher accuracy from the proposed model. The Drop out layer also helped to remove
the non-trainable parameter which helped to reduce the time for training the model.
will be used to classify the images.

The accuracy obtained while training the proposed model is 98.13 percent. The loss
while training the model was 5.66 percent.

44
 CHAPTER

REFERENCES

[1] Mishra, S., Sachan, R., & Rajpal, D. (2020). Deep Convolutional Neural Network
based Detection System for Real-time Corn Plant Disease Recognition. Procedia
Computer Science, 167.
[2] Anjna, Sood, M., & Singh, P. K. (2020). Hybrid System for Detection and
Classification of Plant Disease Using Qualitative Texture Features Analysis.
Procedia Computer Science, 167(2019).
[3] Agarwal, M., Singh, A., Arjaria, S., Sinha, A., & Gupta, S. (2020). ToLeD:
Tomato Leaf Disease Detection using Convolution Neural Network. Procedia
Computer Science, 167(2019).
[4] Ji, M., Zhang, L., & Wu, Q. (2020). Automatic grape leaf diseases identification
via UnitedModel based on multiple convolutional neural networks. Information
Processing in Agriculture.
[5] Rahman, C. R., Arko, P. S., Ali, M. E., Iqbal Khan, M. A., Apon, S. H., Nowrin,
F., & Wasif, A. (2020). Identification and recognition of rice diseases and pests
using convolutional neural networks. Biosystems Engineering.
[6] Picon, A., Seitz, M., Alvarez-Gila, A., Mohnke, P., Ortiz-Barredo, A., &
Echazarra, J. (2019). Crop conditional Convolutional Neural Networks for
massive multi-crop plant disease classification over cell phone acquired images
taken on real field conditions. Computers and Electronics in Agriculture,
167(September).
[7] Lee, S. H., Goëau, H., Bonnet, P., & Joly, A. (2020). New perspectives on plant
disease characterization based on deep learning. Computers and Electronics in
Agriculture, 170(December 2019).
[8] Chen, J., Chen, J., Zhang, D., Sun, Y., & Nanehkaran, Y. A. (2020). Using deep
transfer learning for image-based plant disease identification. Computers and
Electronics in Agriculture, 173(March).
[9] Sethy, P. K., Barpanda, N. K., Rath, A. K., & Behera, S. K. (2020). Image
Processing Techniques for Diagnosing Rice Plant Disease: A Survey. Procedia
Computer Science, 167(2019).

45
[10] Anjna, Sood, M., & Singh, P. K. (2020). Hybrid System for Detection and
Classification of Plant Disease Using Qualitative Texture Features Analysis.
Procedia Computer Science, 167(2019).
[11] Hu, G., Yang, X., Zhang, Y., & Wan, M. (2019). Identification of tea leaf
diseases by using an improved deep convolutional neural network. Sustainable
Computing: Informatics and Systems.
[12] Agarwal, M., Gupta, S. K., & Biswas, K. K. (2020). Development of Efficient
CNN model for Tomato crop disease identification. Sustainable Computing:
Informatics and Systems.
[13] Arnal Barbedo, J. G. (2019). Plant disease identification from individual lesions
and spots using deep learning. Biosystems Engineering, 180(2016).
[14] Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using deep learning for
image-based plant disease detection. Frontiers in Plant Science, 7(September),
[15] Kaur, S., Pandey, S., & Goel, S. (2018). Semi-automatic leaf disease detection
and classification system for soybean culture. IET Image Processing, 12(6).
[16] KC, K., Yin, Z., Wu, M., & Wu, Z. (2019). Depthwise separable convolution
architectures for plant disease classification. Computers and Electronics in
Agriculture, 165(December 2018).
[17] Khan, M. A., Akram, T., Sharif, M., Awais, M., Javed, K., Ali, H., & Saba, T.
(2018). CCDF: Automatic system for segmentation and recognition of fruit crops
diseases based on correlation coefficient and deep CNN features. Computers and
Electronics in Agriculture, 155(October).
[18] Sethy, P. K., Barpanda, N. K., Rath, A. K., & Behera, S. K. (2020). Deep feature
based rice leaf disease identification using support vector machine. Computers
and Electronics in Agriculture, 175(May).
[19] Arsenovic, M., Karanovic, M., Sladojevic, S., Anderla, A., & Stefanovic, D.
(2019). Solving current limitations of deep learning based approaches for plant
disease detection.
[20] Too, E. C., Yujian, L., Njuki, S., & Yingchun, L. (2019). A comparative study of
fine-tuning deep learning models for plant disease identification. Computers and
Electronics in Agriculture.
[21] Zhang, S., Zhang, S., Zhang, C., Wang, X., & Shi, Y. (2019). Cucumber leaf
disease identification with global pooling dilated convolutional neural network.
Computers and Electronics in Agriculture.

46
[22] Sholihati, R. A., Sulistijono, I. A., Risnumawan, A., & Kusumawati, E.
(2020). Potato Leaf Disease Classification Using Deep Learning Approach. IES
2020 - International Electronics Symposium: The Role of Autonomous and
Intelligent Systems for Human Life and Comfort, 392–397.
[23] Madhulatha, G., & Ramadevi, O. (2020). Recognition of plant diseases using
convolutional neural network. Proceedings of the 4th International Conference on
IoT in Social, Mobile, Analytics and Cloud, ISMAC 2020, 738–743.
[24] Gayathri, S., Wise, D. C. J. W., Shamini, P. B., & Muthukumaran, N. (2020).
Image Analysis and Detection of Tea Leaf Disease using Deep Learning.
Proceedings of the International Conference on Electronics and Sustainable
Communication Systems, ICESC 2020, Icesc, 398–403.
[25] Jasim, M. A., & Al-Tuwaijari, J. M. (2020). Plant Leaf Diseases Detection and
Classification Using Image Processing and Deep Learning Techniques.
Proceedings of the 2020 International Conference on Computer Science and
Software Engineering, CSASE 2020, 259–265.
[26] Ashok, S., Kishore, G., Rajesh, V., Suchitra, S., Gino Sophia, S. G., &
Pavithra, B. (2020). Tomato leaf disease detection using deep learning techniques.
Proceedings of the 5th International Conference on Communication and
Electronics Systems, ICCES 2020, Icces, 979–983.
[27] Jiang, Di., Li, F., Yang, Y., & Yu, S. (2020). A Tomato Leaf Diseases
Classification Method Based on Deep Learning. Proceedings of the 32nd Chinese
Control and Decision Conference, CCDC 2020, 1446–1450.
[28] Mathulaprangsan, S., Lanthong, K., Jetpipattanapong, D., Sateanpattanakul,
S., & Patarapuwadol, S. (2020). Rice Diseases Recognition Using Effective Deep
Learning Models. 2020 Joint International Conference on Digital Arts, Media and
Technology with ECTI Northern Section Conference on Electrical, Electronics,
Computer and Telecommunications Engineering, ECTI DAMT and NCON 2020.
[29] Biswas, R., Basu, A., Nandy, A., Deb, A., Chowdhury, R., & Chanda, D.
(2020). Identification of Pathological Disease in Plants using Deep Neural
Networks - Powered by Intel® Distribution of OpenVINOTM Toolkit. Indo -
Taiwan 2nd International Conference on Computing, Analytics and Networks,
Indo-Taiwan ICAN 2020 - Proceedings, 45–48.
[30]Shrivastava, V. K., Pradhan, M. K., Minz, S., & Thakur, M. P. (2019). Rice plant
disease classification using transfer learning of deep convolution neural network.

47
International Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences - ISPRS Archives, 42(3/W6), 631–635.

48