(Project Id: 2020CSEAI001)

DETECTION OF COVID-19 WITH CHEST X-

RAY
A Project Report Submitted
In Partial Fulfillment of the Requirements
for the Degree of

BACHELOR OF TECHNOLOGY
in
COMPUTER SCIENCE AND ENGINEERING - AI
by

STUDENT NAME (ROLL NO. 1816510501)

STUDENT NAME (ROLL NO. 1816510501)
STUDENT NAME (ROLL NO. 1816510501)

Under the Supervision of

Dr. Vineet Kumar Singh
Assistant Professor
Department of Computer Science & Engineering - AI
ABES INSTITUTE OF TECHNOLOGY, GHAZIABAD

AFFILIATED TO
DR. A.P.J. ABDUL KALAM TECHNICAL UNIVERSITY
(Formerly Uttar Pradesh Technical University, Lucknow)
UTTAR PRADESH, LUCKNOW
June, 2024
 DECLARATION

I hereby declare that the work presented in this report entitled “Detection of Covid-
19 with Chest X-Ray", was carried out by me. I have not submitted the matter
embodied in this report for the award of any other degree or diploma of any other
University or Institute.
I have given due credit to the original authors/sources for all the words, ideas,
diagrams, graphics, computer programs, experiments, results, that are not my original
contribution. I have used quotation marks to identify verbatim sentences and given
credit to the original authors/sources.
I affirm that no portion of my work is plagiarized, and the experiments and results
reported in the report are not manipulated. In the event of a complaint of plagiarism
and the manipulation of the experiments and results, I shall be fully responsible and
answerable.

(Candidate Signature)
Name : Name of the student 1
Roll. No. : 2119000000
Date : 06/04/2024

(Candidate Signature)
Name : Name of the student 2
Roll. No. : 2119000000
Date : 06/04/2024

(Candidate Signature)
Name : Name of the student 3
Roll. No. : 2119000000
Date : 06/04/2024

ii
 CERTIFICATE

Certified that Rajesh Kumar (Roll no. 211900000) has carried out the research work
presented in this Project entitled “Detection of Covid-19 with Chest X-Ray” for the
award of Bachelor of Technology from Dr. APJ Abdul Kalam Technical University,
Lucknow under my/our supervision. The Project embodies results of original work,
and studies are carried out by the student himself/herself and the contents of the
Project do not form the basis for the award of any other degree to the candidate or to
anybody else from this or any other University/Institution.

Signature Signature

Dr. Vineet Kumar Singh Dr. Sanjeev Kumar

Assistant Professor Professor & Head(CSE-AI)
Department of CSE-AI Department of CSE-AI
ABES Institute of Technology, ABES Institute of Technology,
Ghaziabad Ghaziabad
Date: 06April, 2024 Date: 06 April, 2024

iii
 ABSTRACT
Currently, agriculture has become the most significant approach than it has been used
before few years back, when the plants are used for feeding the human and flora and
fauna. Presently, the plants have been used to create the electricity and other type of the
power to enhance the living situations of the social beings. Due this, it is requirement for
the suitable care of the plants to acquire the maximum advantage. The main region that
required main consideration is the cutting plant leaves diseases. Number of the diseases
affects the leaves that may cause maximum destruction to different economic and social
aspect. In proposed system, study the different leaf infections using detection and
classification technique in image processing. Initially, various paddy leaves acquire using
digital pictures. After that, RGB model converted into the HSV model for resizing the
picture using k mean clustering with image segmentation. Then, the specific features are
extracted using PCA algorithm. Moreover, the feature extraction and BFO-DNN method
implemented for classification of the paddy leaf diseases. This classification method is
used to improve the detection rate and reduce the entropy loss. It is highly efficient and
accurate to detect or recognize the disease image with different number of categories
(Bacteria light, sheath rot, Brown spot and Normal etc). Experimental analysis is done to
calculate the performance metric like as accuracy, TPR, TNR, FDR, Cross Entropy and
FPR. Then, the comparative analysis of the existing parameters compared to the current
parameters.

iv
 ACKNOWLEDGEMENT
With deep gratitude I express my earnest thanks to my esteemed supervisor Dr. Vineet
Kumar Singh, Assistant Professor, Department of Computer Science & Engineering -
AI for his constant involvement, energetic efforts and proficient guidance, which gave
me direction and body to work, respond here. Without his counsel and encouragement,
it would have been impossible to complete the Project work in this manner.

I wish to express my gratitude to Dr. Sanjeev Kumar (Head of Department), Dr. M. K.

Jha (Director), for their support, guidance and advice throughout this work. I am
thankful to all the faculty members of the Computer science and Engineering-AI
Department especially for their intellectual support during my research work.

I also want to thank my friends for their valuable support whenever I needed. I would
like to thank all those people who have helped me some way or the other in my Project
work.

Lastly, and most importantly, I thank my parents for their moral support and
encouragement towards completing my Project successfully. In the last, I want to thank
Almighty God.

Signature :
Name : Name of the student 1
Roll. No. : 2119000000
Date :
Place :

Signature :
Name : Name of the student 2
Roll. No. : 2119000000
Date :
Place :

Signature :
Name : Name of the student 3
Roll. No. : 2119000000
Date :
Place :

v
 TABLE OF CONTENTS
Description Page No.
Declaration ii
Certificate iii
Abstract iv
Acknowledgement v
Table of Contents vi-vii
List of Tables viii
List of Figures ix
List of Abbreviations x
Chapter 1: Introduction 1-20
1.1 Leaf Disease 1
1.2 Disease Symptoms 1
1.2.1 Symptom of Viral Disease 2
1.2.2 Symptom of Bacteria 2
1.2.3 Symptom of the Fungi 3
1.3 Paddy Leaf Disease 3
1.4 Types of Paddy Leaves 3
1.5 Detection Methods of Paddy Leaf Disease 5
1.5.1 Acquisition of Image 6
1.5.2 Image Pre-processing 6
1.5.3 Segmentation of Image 6
1.5.4 Extraction of the Features 7
1.5.5 Classify and Detect the Disease 7
1.6 Existing Methods: Jaya Optimization Algorithm 7
1.7 Paddy Leaves Colour Model 7
1.7.1 HSV Model 7
1.8 Image Segmentation Using K-Means Clustering Method 8
1.9 Proposed Methods Using PCA, BFOA-DNN Classification 11
1.9.1 PCA Algorithm Used for Feature Extraction 11
1.9.2 BFOA Using Instance Selection (Bacteria Foraging 14

vi
 Optimization Algorithm)
1.9.3 Deep neural Network (DNN) Used for Classification 17
Chapter 2: Literature Survey 20-29
Chapter 3: Present Work 30-35
3.1 Problem Statement 30
3.2 Research Gaps 30
3.3 Research Objective 31
3.4 Research Methodology 31
Chapter 4: Result and Discussions 36-45
4.1 Performance Parameters 36
4.2 Dataset Used in Paddy Leave Images 37
4.3 Experimental Analysis 37
Chapter 5: Conclusion and Future Scope 46-47
5.1 Conclusion 46
5.2 Future Scope 47
References 48-51
List of Publications 52
Plagiarism Report 53

vii
 LIST OF TABLES
Table 2.1 Analysis Several Methods, Merits and Performance Metrics in 28
Paddy Leave Disease Detection System
Table 4.1 Proposed Parameter with Hybrid BFOA-DNN Method 48
Table 4.2 Existing Performance 48
Table 4.3 Comparison: Proposed and Existing Models (Hybrid, 49
DNN+JAO, DNN)
Table VII.1 Conversion of Data Types Commands 61

viii
 List of Figures

Fig. 1.1 Viral Symptoms 2

Fig. 1.2 Bacterial Symptom 2
Fig. 1.3 Fungi 3
Fig. 1.4 Leaf Blast 4
Fig. 1.5 Brown Spot 5
Fig. 1.6 Deep Brown Spot 5
Fig. 1.7 Major Concept of Leaf Disease Detection 6
Fig. 1.8 HSV Colour Model 8
Fig. 1.9 Flowchart of K Mean 10
Fig. 1.10 The Feature Extraction Flow Diagram, Using PCA Algorithm 13
Fig. 1.11 Instance Selection Using BFOA algorithm 17
Fig. 1.12 Architecture of DNN 18
Fig. 1.13 Flowchart of DNN 19
Fig. 3.1 Main Proposed Flow Chart 35
Fig. 3.2 Input Image 36
Fig. 3.3 Filtered Image 36
Fig. 3.4 Segmented Paddy Leave Image 37
Fig. 3.5 (i) Feature Extraction 37
Fig. 3.5 (ii) Feature Selection 37
Fig. 4.1 Paddy Leaf Dataset Images 40
Fig. 4.1 (a) Bacterial Blight 40
Fig. 4.1 (b) Brown Spot 40
Fig. 4.1 (c) Blast 40
Fig. 4.2 Sample Paddy Images 40
Fig. 4.3 Main Project Desktop Application Window 41
Fig. 4.4 Upload Paddy Leave test image, Gray image, HSV image, 42
Noise image, and Filter Image
Fig. 4.5 Feature Extraction and Selection 43
Fig. 4.6 Category Recognize and Classification Image 43
Fig. 4.7 Accuracy with Proposed Method Using Hybrid BFO- DNN 44
Fig. 4.8 False Discovery Rate with Proposed Method Using Hybrid 44
BFO- DNN
Fig. 4.9 Cross Entropy Loss with Proposed Method Using Hybrid 45
BFO- DNN
Fig. 4.10 True Positive Rate with Proposed Method Using Hybrid BFO- 46
DNN
Fig. 4.11 True Negative Rate with Proposed Method Using Hybrid 46
BFO- DNN
Fig. 4.12 Comparison Analysis with Accuracy Rate (%) 47
Fig. 4.13 Comparative Analysis with Entropy Loss (%) 47
Fig.V.1 Editor Window 60
Fig.VI.1 Workspace 60
Fig.VIII.1 Blank screen GUIDE in GUI 61

ix
 List of Abbreviations

PLD Paddy Leave Detection

DIP Digital Image Processing
BBL Bacteria Blight Leave
SRL Sheath Rot Leave
BSL Brown Spot Leave
MATLAB Matrix Laboratory
GUI Graphical User Interface
GUIDE Graphical User Interface Development Environment
WS Workspace
BFOA Bacterial Forging Optimization Algorithm
PCA Principal Component Analysis
FS Feature Selection
DNN Deep Neural Network
JOA Jaya Optimization Algorithm
TPR True Positive Rate
TNR True Negative Rate
FDR False Discovery Rate
FPR False Positive Rate
EL Entropy Loss
KMC K-means Clustering
CI Cluster Index
HSV Hue, Saturation and Value
RGB Red, Green, and Blue
IB Image Binarization
AUC Accuracy Rate
GSI Gray Scale Image
3D-Box Three Dimensional -Box Filtration
GF Guassian Filter
FPs Firefly Positions
CW Command Window
EW Editor Window

x
 CHAPTER 1

INTRODUCTION

In early the early stage, India is the fast-developed country and agriculture is
the back-bone for growth of the countries. Generally, the obstacles are faced because
of the industrialization and globalization. One of the essential requirements for the
cultivation to be inspired to the younger-generation. Presently, computerized
technology plays an essential role in the field of the agriculture [1]. The detection of
the plant disease is related to maximum loss of the yield, time, currency and quality
product. Hence, the situation of the plant is necessary for the cultivation of the crop.
In the previous days the identification was done through the manual process, but
because of the environmental variations, the identification of the plant disease
becomes difficult.
1.1 LEAF DISEASES
Generally, the leaves influenced by bacteria, fungi and Virus. The symptoms
of the plant disease are available on the leaf. Some of the symptom are identified by
the change in colour, shape and reason for the response of plant to pathogens [2].
1.2 DISEASE SYMPTOMS
Normally, the diseases of the plant take place in the various phases:
• Fungi, bacteria and Virus leads to infections to plant.
• The acidity and toxic nature may lead to disorder of the plan
• Environmental changes
1.2.1 Symptom of Viral Disease
Generally, the viruses found in the plants are not easy to diagnose.No tell-tale
sign iscreated by the virus, that may be easily identified and often disorder along
with deficiency of the nutrient and herbicide wound. The leaves might be
compacted, frizzy and developed in stunted way.

1
 Fig.1.1 Viral Symptom [3]

1.2.2 Symptom of Bacteria

Normally, the pathogen bacteria are found mainly in the in the vegetable
plants. It is the dangerous kind of the disease [3]. It may not pierce directly in to the
tissue of the plant but it requires to enter by the injury or standard plant initial point.
The injuries are caused due the damage caused by the insects, and other pathogens,
and equipment at the time of the processes like as clipping and harvesting. The small
green spots that are mainly water soaked feature the disease. In this disease, the
lesion is increased in size and after that it appeared as the dry dead spots as given in
Fig. 1.2

Fig. 1.2BacterialSymptom [3]

1.2.3 Symptom of the Fungi

This disease takes place due to the presence of the fungus. For instance, late
blight may occur because of the fungi. It occurs at grown up leaves such as leaves
that are soaked by water, grey-green leaves.

2
 Fig.1.3 Fungi [3]

1.3 PADDY LEAF DISEASE

Generally, paddy leaves are exaggerated by fungi and bacteria infections. The
unrecognised diseases of paddy leaves may origin a major demolition to the
cultivation of the crop and, it results in the less production of the crops [4]. Different
diseases effects on the growth of the paddy and some of the paddy leaves diseases
are listed as,
• Blast
• Bacteria Light
• Brown Spot
• Sheath Rot and
1.4 TYPES OF PADDY LEAVES
The different kind of the paddy leave diseases are[5];
1. Blast:
The major cause of the disease is the fungus magnaporthe oryzae. Generally,
colour of the node and the neck part is affected by this disease.

3
 Fig.1.4 Leaf Blast[5]

Presence
It takes place in all the stages, when there is occurrence of the pores. It takes
place in minimum moisture quantity and temperature.
Symptoms
It takes place in the presence of the green marks with dark green boundary.
Lesion is in the form of the oval and also brown at the middle section.
2. Brown spot:
It is mainly influencing the sheath and seed of the leaf. It damages the
complete leaf.
Presence
The environments with supre memoisture and soil with a smaller amount
nutrient level resulted this disease. The fungi present in the seed may live for four
years.
Symptoms
Seeds comprise yellow-brown lesions at first stage. After that, it converts purple
brown and surrounded by reddish-brown edge.

4
 Fig. 1.5 Brown Spot[5]

3. Deep brown spots

It took place because of the sphaerunlina oryzina fungi. Thismostlypollutes
sheaths of leaves [4].
Presence
It takes place in soils that is deficient in potassium. It takes place at previous
phase of the plants.
Symptoms
It has the lesions that have dark spots and it can be 2-10 mm dimension and 1-1.5
mm wider.

Fig. 1.6 DeepBrown spot[5]

1.5 DETECTION METHODS OF PADDY LEAF DISEASE

Generally, the detection of the plant leaves is determined through the

identification and classification approach. The various stages are explained as:

5
 Acquisition of Image

Image Pre-processing

Segmentation of Image

Extraction of the Features

Classify and detect the disease

Fig. 1.7 Major Concept of Leaf Disease Detection [8]

1.5.1 Acquisition of Image

The initial stage of visualization network is the image acquisition. It includes
the stages to acquire the plant leaf and, captured the high-quality pictures through the
digital camera [6]. The pictures are captured from the social sites or the farming
field. The proficiency is based on the quality of the dataset pictures. The pictures are
in the form of the RGB format.
1.5.2 Image Pre-processing
This includes the stage of the picture improvement, RGB to Lab
transformation, filtration and so forth. However, picture improvement is supported
for improving the contrast picture. And, the smoothening of the image is done by the
filtration method. Various kinds of the filtration methods are image processing such
as median filter, and so forth.
1.5.3 Segmentation of Image
Normally, picture segmentation is the segmentation of the picture that is
divided in to similar features and comprising same features. The segmentation is
done through the otsu technique, k mean clustering approach, transforming the RGB

6
picture into HIS method. K-mean clustering classified the object based on the group
characteristics into k amount of the modules. The objects classified through the
decreasing of the amount of the squared of the displacement among the object and
related cluster[7].
1.5.4 Extraction of the Features
It plays significant role in the detection of the object. When the image
segmentation is done, the extraction of the features is acquired form the picture. In
various applications of the image processing the feature extraction process is utilised.
In case the color features are extracted, then it includes the color, histogram,
moments and design descriptor. One of the techniques for the extraction of the
textual feature is gray level co-occurrence matrix
1.5.5 Classify and Detect the disease
Lastly, the datasets trained and tested the classifiers. Some of the classifiers that is
used to identify and classify disease is SVM, KNN, NN, Fuzzy logic-based system.
1.6 EXISTING METHODS:JAYA OPTIMIZATIONALGORITHM
Jaya optimization is globalized algorithm that has been efficiently acquired to
normal function of limitation and uncontrolled problems. This algorithm may not
need any defined metrics that requires the wider alterations before the generation of
the accurate observations. The directive of this algorithm is not simple method.
Moreover, adapting the parameters in each iteration is not informal and ingests
maximum time.
1.7 PADDY LEAVES COLOR MODEL
There various models available for paddy leaf. But, one of the model is HSV
model that is described below:
1.7.1 HSV model
HSV model determines the single form of the Hue, Saturation and Brightness
rate. Hue is related in direct way to the notion of Hue in the color major portion [18].
Some of the advantages of the Hue are:
• The relation among the tones towards the color circle is simply recognised.
• The colors, shades and tones may be created in absence without the influence
of the Hue saturation related directly to the notion of the tint in the color part,

7
 instead of the complete saturation does not create the tint, whereas zero
saturation creates white, black and grey shade.
The advantage of HSV is the every of the characteristics is related in direct way to
the major color ideas, that makes it easy theoretically. In HSV model, the color
scheme determines the hue shift, saturation and value that is called as HSV.

Fig. 1.8. HSV Colour Model

1.8 IMAGE SEGMENTATION USING K-MEANS CLUSTERING

METHOD

The segmentation based on the analysis of the cluster is called as the k mean
clustering. This technique is used to solve the low-level segmentation procedure.
Initially, the clustering procedure is used for the segmentation of the large-scale
images. Clustering is the unsupervised learning method that is the set of the features
which is interconnected in the non-changing clusters. K mean clustering is the
segmentation of the collected data in to the k amount of the data. It categorizes the
set of the data into k –segment clusters [20]. K-mean consists of the two steps,
which calculate the k centroid at the starting stage and each point to set worked with
the nearest centroid from the specific data points. Different methods demonstrated
the distance of the nearest centroid and it is mainly used methods that are Euclidian
distance. In case the clustering is processed then computing is done again if the new
centroid of each cluster and it is based on the centroid. The Euclidian distance is
calculated between each centroid and data points aligned the set points that have the
minimum Euclidian-distance. Each set-in segmentation is described as the set objects
and, moved through the centroid. The centroid cluster is the notion when the quantity
of the distance from each object in cluster gets reduced. K-mean is the selection

8
algorithm that reduces the quantity of the areas from each object to the set of the
centroid.
Assume the image with the resolution of x X and the image consists the set with k
groups and clusters. Let (z,x) is input image as the cluster of each R,G,B network
where j=4 and , is the cluster centroid of the network j with color image. The
algorithm is described as:
• Store the color image.
• Disconnect the image into Red, blue and Green station.
• Initialize the quantity of the cluster k and intermediate network.
• For each pixel of image, calculate the Euclidian distance d, between
the intermediate and each pixel of image, j determines each channel (1.1
and k denotes specified cluster.
=∑ (z,y) -
• Allocating each pixel to nearest centroid that is based on the distance D.
• When the complete pixels have assigned, and again computed then
(1.2
the new position of the centroid of each network by the average
amount of the pixels.
= ∑ ∈ ∑ ∈ (z,y)

• Repeat the procedure until it satisfies the lost rate.

• The cluster pixel gets re-shaped.
It is the process where initial stage is the pre-processing that segments the pictures
into various clusters. The clustering process uses the k mean color based approach.
Then, the various clusters are selected and disease is identified from a unique cluster.
The simple unsupervised learning method resolves the desired clustering issue. The
procedure followed normal method for the classification of desired database through
definite amount of the clusters. The main goal of this algorithm is to define the k
center, each with single cluster. The method placed in cunning way due to diverse
outcome at various positions. However, the suitable selection should be made [19].
The other stage is related to the desired database and connects it to closest center.
When no portion is awaiting, the initial stage is complete and at initial group period
are stored. It is required to recomputed the k centroid of the cluster achieved from
preceding stage. Then, k novel centroid connected among the similar database and

9
closest centers. Loop is created and it may alter k center by modifying the position
one by one unless no modification take place where no center may replace.

Start

Number of K clusters

Centroid

Distance objects of k Clusters

Group depends on minimum Distance

No Object Move No
Group?

Yes

End

Fig. 1.9 Flowchart of K Mean

Pseudo code of K mean clustering algorithm

Step 1: Initialization

Input: The k number of the clusters;

Step 2: Group of K clusters;

Selecting the k object as initial centroid cluster;

Step 3: Repeat the process;

10
Reallocate every object to cluster where objects are same the value of object in
cluster;

Step 4: Updating process;

Updating the average cluster;

Compute the mean value of object for every cluster;

End while

End;

1.9 PROPOSED METHODS USING PCA, BFOA AND DNN

CLASSIFICATION

1.9.1 PCA Algorithm Used for Feature Extraction

It is the technique to reduce the dimension of the information. It predicts the
dimensional data to fewer dimensions and projects the data into least four-sided
sense. It includes the main alterations in the data, pays no consideration to minimum
inconsistency, and reduces the dimension of dataset by finding the new variable that
is minimum as comparable to the set of the variables. It may accumulate the main
sample information that eliminates the related information from uncertain dataset [9].
It is not a parametric method for lessening of the dimension. It is used for
compression and classification of the data. It finds the eigen vectors of the co-
variance matrix with highest Eigen value and used data with new similar and
minimum dimension. PCA converts the matrix of n features into new dataset is
minimum than n features. It reduced the quantity of the features by creating a new
and minimum amount of the variables that access the important variables [10].
Stages:The various stages are described as:
1. Select the data
The co-variance matrix is the series of the linear functions that may be
scaling or the rotational. The eigen value segments the transform metric in which the
eigen vector recognises the mobility matrix and the eigen value demonstrates the
scaling-vector. The co-variance matrix may not contain the translation value. Hence,
PCA is acquired to rotate the data to achieve the non-related axes.

11
2. Normalization of data
The eigen vector of the co-variance matrix in the location of variance of data.
However, variance is the exact level of the comprehensive data. The variance of the
data is measured that is maximum than the variance of the same data. When the
information is not normalized, then the highest variance is the eigen vectors. For
preventing the scale dependent nature of the PCA, it is significant to normalize the
data by disseminating each feature by the standard deviation.
3. Calculate the Eigen Distribution
When the data is probable in to the highest eigen vectors to reduce the
dimensionality, then the eigen fragmentation can be acquired. Mainly, the
fragmentation is acquired by unique value.

4. Planning the data

The data is projected in to the highest eigen vector for decreasing the
dimension. Let the matrix V where the column contains the eigen vectors and D is
the real data that contains the variant clarifications. And then, the projected data is
acquired by V ¬ D.

∑
T=∑ (1.3

However, the selection of the quantity of the remained dimension, that defined the
amount of the variance of real data.

In case, S is the eigen vector that are stored and ….. recognises the interrelated
eigen values, and the amount of variance may stores after the planning of the
accurate dimension is calculated as:

12
 Input data matrix

Decrease in data dimension

Result of Principal Components

Principal Components input scheme

Output the value

Fig.1.10 The Feature Extraction Flow diagram, using PCA algorithm [11].

Advantages ofPCA:

• Decrease the complexity of set of the pictures by using PCA.

• Minimum dataset demonstrates when the trained pictures are stored in for the
predictions based on decreased basis.

Pseudo code in PCA Feature Extraction Method:

Process PCA

Begin

Create random population of P output (chromosome);

For every individual j P , compute the fitness (j);

For k=1 to # generations

For every individual j P : do j= localised search(j);

For cross-over

13
Selecting the binary parents , P randomly

Generating offspring to P;

=local searching ( );

Adding individual to P

End for

For mutation

Selecting the individual j P randomly;

Create offspring = mutation (j);

= local search ( );

Adding the individual to P;

End for

P = selecting P;

K=k+1;

End for

End

1.9.2 BFOA using Instance Selection (Bacteria Foraging Optimization

Algorithm)
BFOA is theBacterial Foraging Optimization Algorithm. In this method, the
optimization of the bacterial chemo taxis approach. Generally, optimization of the
integrating is based on foraging behaviour of the bacteria [12].BFOA is related to the
bacteria optimization and swarm optimization method used in the computer
intelligence field. This algorithm is inspired by the set of the foraging behaviour of
the bacteria like as E.coli.
Stages of Algorithm:
1. Chemo taxis: This process imagines the motion of E.coli cell by the spinning and
tumbling. Assume the θxj(l, t, m) that determines the jth bacterium at th chemotactic,
mth reproductive, and lth elimination–dispersal. The random position is determined
the length elements.The procedure stimulates the motion of E.coli cell by swimming

14
and tumbling through the flagella. It may swim for the given time interval and
interchange among the two models for the complete lifetime.
2. Swarming: The integration of various species of the bacteria contains E.coli and
S. typhimurium in which the designs are searched in the swarm mode [13]. The
gathering of the E.coli positions the nutrient slope point along with unique nutrient
chemo-effector. The cells are simulated by the higher value into groups and
maximum density.
3. Reproduction: The less healthy bacteria may die at last because every healthier
bacteria may be asexually segmented into binary bacteria. The bacteria located at the
similar position and swarm dimension remained static.
4. Elimination and Dispersion: The process of BFOA in which few bacterium are
placed at random location with minimum probability, while the new alternates are
randomly located above the search space [21].Sudden modification in the normal
environment, and where, a bacteria population life may take place due to different
reason. Events may take place so that bacterium in area may died is disseminated
into novel portion of the setting.
Advantage of BFOA:

• Simplification of the linear reduced method.

Pseudo code in Instance Selection:

Start
.
Generate the random swarm value bacteria f ( ( j, k, l )) ∨ , j=1,……

For m=1 to ! ." do

For l=1 to !#." do

For k=1 to !$.% do

For j=1 to !&.# do

.'
Updating the valuef ( ( j, k, l )) ∨ , j=1,…… to imitate the swarm method

Performing the chemo-taxis procedure

Tumble Swim process

15
 .
For bacteria f ( (j, I, k)) regulated by !# )

End for

End for

Process the reproduction method by arranging the bacteria that depends on the
swarm,f ( . (j +1 ,I, k))

End for

End

Fig. 1.11 Instance Selection Using BFOA algorithm [14]

16
1.9.3 Deep neural network (DNN) used for classification

Neural network is the interconnection arranged in easy processing

components, hops that whose functions are roughly depend on the animal neuron.
The probable capability of the system is placed in the internal interconnection power
or the weights that is acquired by the procedure of adapting and learning a group of
the trained design.
DNN is the subgroup of the artificial intelligence network that integrated with the
combined hidden layers among the input and output layers. It contains the
complicated non –linear relation models.
The neural networks are intended with diverse pooling and convolution layers. After
that, ANN is established in 1979, that is also known as non-cognition that is utilised
as hierarchical, multiple layer patterns. These patterns permit the computer the
“learn” to determine the visualised designs. DNN is the set of the neurons that is
sequence of the diverse layers where the neurons acquired activated neuron and
performed the computation. The neurons of the network mutually implement the
compound nonlinear mapping through the input to output value. The mapping is
recognised from the data by alignment of the weight of each neuron used through
back propagation [15]. This technique is built by neuron in the maximum layer. The
high-level layer is the non-concrete and internal domain of the DNN that is mainly
interpretable. Deep Learning is related to the neural network structural design that is
related with maximum quantity of the layers where the starting seven layers have
extracted the essential features from the information(Y) and final layers are mainly
interconnected by relating the extracted feature to the output value (Z). In the given
Fig., the DNN is the series of the full connected layers that receive the features of
every element as input and classified the values. In case the input features are getting
DNN, then the output rate are calculated in sequence towards the layers of the
system. At every layer, weigh average is computed by productive value productive
rate of input vector consisting the output rate of every unit in last layer by weigh
vector for every component in the present hidden layer. The prototype rate is created
by maximum activated value. DNN is more capable of learning the higher-level
features with the highest complexity and simplication is linked to area neural
network [16].

17
 Fig.1.12 Architecture of DNN [17]

Collecting the data

Trained data Testing data

Feature Extraction

Deep neural network (Training)

Deep neural network (Testing)

No
If model is satisfied

Yes

Compute and compare the value

Prediction Result

Fig. 1.13 Flowchart of DNN [17]

18
DNN is commonly preceding the vision of the data. The pixels of the images are
located into the first layer of DNN and output layer is built as the demonstration of
the capability of the lower level features of the image. The features are combined as
the succeeding layers with higher level features. The given Fig. demonstrates the
vision picture of the calculation NN. Few of the values are acquired by the input
layer and broadcast them to neurons in the medium layer of the network that is
known as the hidden layer. The weigh amount with high amount of the hidden layer
is mainly computed to the out layer that offered the final value of the network to
user. The output neurons termed as the activations and synapse derived as weights
[22].

Advantage of DNN:

• In Language Modelling
• Computerized Vision

Pseudo Code for classification Method:

Stage 1: Initialize the value Y=(( , () ,…….,( )that represent the data-matrix with n
number of the samples, and Z=(* , *) ,…….,* )" because the related output marks,
the highest amount of the selected attributes k.

Stage 2: Initialization: T = {bias} , D =E and , = 0

Stage 3: whereas, ∥ T∥≤ l+1 do

Stage 4: Allocate , (candidate weigh) = 0;

Stage 5: Updating weigh of hidden layers and also the input weight ,/ ;

Stage 6: Produce the drop time out to be utilized and achieve average 01.2

k = 345 62 max ∥ 01.2 ∥8

Stage 7: Compute the value

Stage 8: Update learning value using Ada Delta

Stage 9: Initialization ,1. along with initialization

Stage 10: Performing 9 = 9 ∪ < / /'"=

> >/@A

19
 CHAPTER 2

LITERATURE SURVEY

2.1 LITERATURE SURVEY

Ramesh, S et al., 2019 [23] proposed research on the detection and grouping
of the paddy leaf diseases through optimized DNN (deep neural network) along with
java procedure. Using picture acquisition, the pictures of rice-plant-leaves were
mainly caught from the farmhouse playing field for standard, bacteria blight, brown
spot and blast diseases. During pre-processing, for contextual elimination the RGB
pictures were transformed into HSV pictures and that depends on Hue(H) and
(S)Saturation-part binary pictures were eliminated to segment diseases and non-
diseased parts. Moreover, for the separation of the infected part, standard and
background, the clustering approach was used. Infected leaves were classified
through optimized deep-neural network (DNN-JOA) along with java optimized
algorithm. To define the static nature of this algorithm, response loop was produced
in post-processing stage. Investigational outcome was computed through ANN, DAE
and DNN. The projected algorithm has attained accuracy up to 99% for the
exaggerated blast, 96% for bacterial -blight, 94% for brown –spot and for standard
leaves picture was 91%.
Prajapati, H. B et al., 2017 [24] presented a sample network for recognition and
ordering of the rice infections that depends on diseased rice plant pictures. The
prototype network was established after the comprehensive survey of different
method used in digital image processing models. They measured 3-rice plant
infections namely bacteria leaves blight, brown spot, leaf smut. They captured the
pictures of the diseased rice plant through digitalized camera from the rice ground.
They numerically compute four methods of the contextual elimination and three
methods for segmentation. They planned a centroid feed-based k mean cluster
approach to enable exact extraction of features and for segmentation of the infected
parts from leaf pictures. In addition, they extracted the features below various
classifiers namely: color, shape and texture. In addition, they used SVM (support
vector machine) for the classification of multiple classes. Moreover, they acquired

20
accuracy up to 93.3% on trained dataset and for the test dataset it was 73.2%. In
addition, they performed five-cross validation with 84% accuracy and ten-fold
validations it was 89%.
Ramesh, S et al., 2018 [25] developed ML algorithm to search the signs of the
disease in rice-plant. An automatic detection of the plant disease was examined using
machine learning method. Normally, the pictures of standard and infected disease
were observed in planned method. There was extraction of the features for the
standard and infected diseased leaves of the rice plant. In addition, the complete
dataset contains 300 pictures and, segmented for training and testing applications.
These pictures were administrated along with planned approach and leaves were
classified as diseased and non-diseased. Simulation outcomes presented the accuracy
about 99% for the blast diseases pictures and 100% for standard pictures at the
training stage. The testing stage accuracy was 90.4 % for infected picture and for
healthy picture it was 86.3%.
Shah, J. P et al., 2016 [26] proposed research on the assessment of different method
but also described the significant notions of IP (image-processing) and MP(machine
learning)that was analysed for recognition and grouping of the plant infections. They
complete point by point investigation of nineteen documents, cover the effort on rice-
plant ailments and another various plant life and organic products, and also current a
study of these papers’ dependent on significant standards. These standards
incorporate dimensions of picture database, amount of the group of objects (list of
diseases), pre-processing, division systems, kinds of classifiers, precision of
classifiers and so on. They overviewed and studied to propose and structure on
location and order of rice plant sicknesses.
Narmadha, R. P et al., 2017 [27] proposed research on the recognition of the paddy
leaves infections. Few of the paddy infections was (blast-disease) BD,(Brown-spot
disease)BSD, (Deep-brown-spot) DBS, that breaks the development and prevention
of paddy. Infections may blight the paddy leaves at various stages of development
and every portion of plants as leaf neck and node. The proposed model was intended
to eliminate the distortion automatically, fault by human, and decreasing the time
interval that influenced the paddy leaf infections. Moreover, it enhances the accuracy

21
rate. In this research, they surveyed k mean algorithm for paddy leaf recognition and
identification.
Islam, T et al., 2018 [28] presented new method to recognise and categorise the
infection based the proportionality of RGB rate of the exaggerated part by IP (image
–processing). If the proportion of the RGB exaggerated area was removed and
assembled into different classes, that were fed to modest classifier named as
NB(Naïve -Bayes) that categorise the infections in to various classes. The method
was positively recognised and, noticed three rice infections namely rice-brown spot,
rice bacteria spot, and rice-blast. The method was effective and fast due to usage of
single feature where RGB rate of diseased part was acquired less calculation time to
recognise and categorise the infections. Instead of handing out the complete leaf, the
model eventually identifies the disease using minimum model of leaf covering the
exaggerated part of the rice sickness.
Singh, A et a., 2015 [29] implemented a new model for recognition of color of
paddy crop vegetable in orientation to LCC (leaf –color-chart), to prevent from some
crop sickness. In this research, the proposed model was recycled to link the crop-leaf
-color to LCC (leaf-color -chart) that have been presented for complete data of plant.
By utilizing digital image processing method, adeveloped and influential technique
for the covering expectation of paddy crop plant had demonstrated andnumerical
representing an incredible phase to the threatening Fig. in the horticulture arena.
Singh, A et al., 2018 [30] proposed research on approach worked on issue of blast
infection of paddy leaves by observing the picture of plant sheet by specialists related
to essential actions. The foundation of the infection recognition algorithm was
Shading Cutting Method, which observes the infected-spots and harmed extent of
complete leaf, building it simple to acquire exhortation if illness exists and dispose of
it inside time in order to stay away from misfortunes. In this research, new technique
was presented for the recognition of the blast-illness of paddy trees. The propose
method compared to existing method to achieve the accuracy up to 98.9%.
Zhou, L et al., 2016 [31] motivated to develop a secure, accurate, and effective
protocols that applied for eigen decomposition and single rate decomposition for
contracting out the ED and SVD. So, as to accomplish security, they utilized
productive protection saving changes to ensure both the information and yield

22
security. So, as to check the accuracy of the outcome came back from the cloud, a
productive confirms conventions were exceptionally effective. They presented a re-
appropriating guideline part examination as a use of our two proposed conventions.
Pinki, F. T et al., 2017 [32] developed an automatic technique that was presented for
analysis of three(3) paddy plants infections namely(brown-spot (BS),leaf-spot(LS),
and bacterial-blight(BB)) and pesticides that were considered in accordance to
strictness of infections. Generally, k mean clustering algorithm was utilised for
differentiating defected part from paddy leaf picture. The kind of the paddy leaf
infections was identified by SVM classifier. After the recognition process, the
prognostic medication was recommended, that helped farming related persons and
co-operations to take suitable action against the diseases.
Suresha, M et al., 2017 [33] classified paddy plants ailments such as BS and BB
that was conceded. Moreover, RGB pictures were transformed into HSV color
picture for differentiation. KNN classifier was used for classification along with
numerical feature such as area, main-axis, small-axis, and boundary. The plant
diseases were approximately categorised into different types which were bacteria,
fungi and virus infections. In planned method, fungal infections were measured. In
this research, they planned model for recognition of BB and BS infections. Global
threshold technique has been analysed and KNN classifier have been utilised to
categorize information. The proposed model achieved accuracy up to 76.6%
Gu, X et al., 2013 [34] introduced a Beer Lambert Law (BLL) to progress the
conversion method of paddy leaves.Generally, the precision of reversal method was
assessed inside-situ tests, comprising coefficient of assurance (R2), RMSE and
generally speaking exactness, while appearing differently in relation to the method of
particular value and multiple-factors. The examination demonstrated that the reversal
method dependent on BLL arrived at most elevated precision than particular value
and multiple-factors methods. The correctness of multi-variantreplicaswas
commonly steadier comparable to unique factor. Be that as it may, the more
exactness of particularflexible method performed better comparable to multi factors
model.
Kapilya G, et al., 2019 [35] summarized various kinds of the plants, categories and
normal methods that helped in achieving the data about computerized visualization

23
and benefits on plant sickness detection and grouping. The presentation analysis of
the various- classifiers was identified and, classified for better representation that,
classically worked between various plant sets and different threats. Moreover, they
compared various group of approaches utilised for testing various group of plants
and diseases. It was analysed from the observation that BPNN classifier have better
performance as compared to other methods such as KNN, SVM, Bayesian and
Feature based algorithms. On other hand, models of diverse plant sets were tested
through diverse approaches from which KNN was observed with better performance
rate.
Mohan, K. J et al., 2016 [36] developed network for the classification of the paddy
plant sicknesses utilising digital picture processing methods. The main goal of the
research was to implement an image processing model that may detect and categorise
different paddy plant infections influencing the farming of paddy namely BS
infection, LB infection and BB sickness. This research was distributed into two main
parts as; paddy plant disease recognition and detection of the paddy plant infections.
During the detection of the disease, the parts of the paddy leaves were recognised
through Haar based features and Ada Boost classifier. The recognition accuracy rate
was found as 83.4%. Moreover, at the time of the recognition of the leave, the paddy
plant category was detected through SIFT feature and classifier namely KNN and
SVM. Using these classifiers, one may identify a previous stage and may acquire
essential stages to reduce the loss of construction. The recognition accuracy of
disease for SVM was 91.10% and KNN was 93.4%.
Reddy, K. N., 2017 [37] surveyed research on the various classification method that
may be used for the grouping of the plant-leaf(PL)illnesses. Generally, identification
of the infections was not easy for the agriculturalists. Therefore, crop prevention was
done using digital image processing method may detect the disease through color
data of the leaves. Various classification methods like as KNN, Probable neural
network, Genetic algorithm, PCA, ANN, and Fuzzy Logic. The classification
technique was not easy to detect that may change for varied input data. This research
mainly focused on the detailed approach of various classification methods utilised for
organization of the PL disease.

24
Arjunagi S et al., 2017 [38] evaluated and condensed systems of the digital image
processing that have been utilized in identification of leaves sickness. They set up
that removal of illness district from the plant sheet picture is the pouring
improvement, and they had considered and looked at different division systems.
They used review and study, proposal of work in a similar way. The paper
introduced an itemized schematic outline of the proposed work and examined
significant advances.
Nidhis, A. D et al., 2019 [39] implemented a digital image processing technique to
classify kind of the affected illness and compute the harshness of the disease by
computing the grade of the affected region, that may be used in regulating the high
level of the pesticides. The disease was the major described diseases which took
place in paddy crops between other diseases mainly affecting the profit and financial
aspect of the farmers.
Kahar, M. A. et al., 2015 [40] presented a combined technique for identifying
infections on paddy plant leaves and presents the user with references on way tp
prevent and regulate diseases. This research mainly focused on three major infections
of paddy in Malaysia that affects the production of the crops namely, bacterial leaf
blight (BLB), blast disease and sheath blight (SB). The technique of detection used in
this research was neuro-fuzzy expert scheme. This technique links the learning
abilities of ANN with social like knowledge demonstration and rectification
capabilities of fuzzy logic schemes and rule based expert schemes. In this research,
the prototype was established to assist Malaysian paddy farers and paddy
investigators by presenting an early detection of the infection in paddy crops and
supporting with crop maintenance. The efficient crop maintenance was essential to
assure the health of crop and receive maximum quality of the productivity. The
recognition accuracy was about 74.21%.
Barbedo, J. G. A. et al., 2013 [41] presented research on techniques that used digital
IP method to identify and categorize plant infections from digitalized pictures in
noticeablerange. Though illness signs may evident in some segment of the plant,
some techniques may discover the display signs in plants and branches that were
measured. The main motives: to restrict the measurement of research and techniques
were related tp roots, seeds, and fruits that similar individualities that may alert the

25
specific investigation. The nominated method was distributed in to three classes in
accordance to the goals: identification, severe quantification and classification. Every
class were again divided in accordance to the major technical output that was utilized
in the approach. This research was estimated to be valuable to investigators
processing on plant pathology and design detection, presenting a complete indication
of the essential arena of investigation.
Anami, B. S et al., 2020 [42] proposed research on the designing the deep CNN
framework for automated detection and classification of different biotic and abiotic
paddy crop stress using field pictures. The research acquired pre-training VGG-16
CNN method for an automated classification of strained paddy crop pictures captured
at the booting development phase. The trained method acquires the accuracy up
92.76% on the stable database representing the technological feasibility of using
deep learning method. It used 30,000 field pictures of 5 various paddy crop variety
along with 12 various stress classifiers. The planned research searches the
applications in increasing the decision support schemes and mobile services for
automating the field crop and resource maintenance services.
Rahman, C. R. et al., 2020 [43] presented the provision of the CNN picture
classification, deep learning models for the detection of the diseases and pest from
rice plant pictures. The main goal of this research was state of art scale structure like
the one as VGG 16 pests and inception V3 have acquired for detection of the rice
infections. Numerical analysis showed the efficiency of the method with the actual
dataset. Though, large scale structure is not appropriate for mobile sensors, the
binary stage CNN structure have planned and compared the state of art CNN
structure like as Mobile Net, Squeeze net. The planned structure acquired the
accuracy 93.3% that was essentially reduced method.
Sharma, R. et al., 2020 [44] developed a possible method using CNN for the
classification and forecast of the infections in paddy crop. Paddy crop was deadly
and it may affect the crop strictly if it was not considered at the starting phase. The
planned method may enhance the decision making using the CNN in case of different
disease in paddy crop for forecast of disease at starting phase and avoiding the
weight loss in production of the complete yield. The model was major for the
categorization and prediction of paddy crop infection and it trained the better quality

26
of the crops. They used CNN method that was costly in calculation so that the
machine with high GPU and processor has maximum dataset of pictures in way that
was more efficient.
Das, S. et al., 2020[45] presented research on the growth of an automated scheme
that analysed the diseased paddy leaf pictures and offer guidance to the farmers. The
main objective of this research was the growth of the rice infected classification
scheme that detects and classified the infected rice in automated way. This research
segments in two sections, namely rice disease identification and classification of the
infected rice. During the identification of the leaf disease, the features were
accountable for the extraction of the infected parts of rice and related features were
acquired using planned model. The planned technique provides significant outcomes
that helped in farming field. This research searches the probability
Rautaray, S. S et al., 2020 [46] proposed research on the detection of the dataset.
The method presented a method for the detection of the infection of lady leaf the
planned model used transfer learning method that was pattern of solving deep
learning issues in an effective way. The method searches the possibility of the
existence of the disease that helped for doing the significant decision related to
specific plant length. The planned method used VGG-16 structure of transfer
learning for the prediction of paddy leaves. The developed method that followed the
detection and collection issue
In table 1 shows the various methods, advantages, issues and parameters have been
described after surveying various papers.

Table 2.1: Analysis Several Methods, Merits and Performance Metrics in Paddy
Leave Disease Detection System

Author Year Technique Advantages Issues Parameters

Ramesh, S 2019 DNN with More Expert Accuracy

et al., [23] JOA consistency advice is
costly.
Prajapati, 2017 K-means Medial filter Lack of Threshold
H. B et al., Clustering remove noise continue
[24] monitoring
Ramesh, S 2018 Machine Automated Economic Accuracy

27
et al., [25] Learning detection loss of
farmer
Shah, J. P 2016 Machine Identify rice - Accuracy
et al., [26] Learning plant disease
Narmadha, 2017 Filtration Remove noise Abnormal Accuracy
R. P et al., automatically functioning
[27]
Islam, T et 2018 SVM Detect small Use of black Classificatio
al., [28] affected area of pixel n accuracy
leaf
Singh, A et 2015 Color Automatic color Atomisation Accuracy
a., [29] Chart- prediction of crop
Model paddy crop health
plant
Singh, A et 2018 Color Identify disease Environment Accuracy,
al., [30] Slicing spots al changes sensitivity
method specificity

Zhou, L et 2016 Eigen de- Detection and - -

al., [31] compositio classification
n model
Pinki, F. T 2017[ SVM Classify disease Inappropriate Accuracy
et al., [32] detection
Suresha, 2017 KNN Classify data Difficult Accuracy
M et al., Classifier identify
[33] brown spot
Gu, X et 2013 Beer- Improve - Accuracy
al., [34] Lambert accuracy
law
Kapilya G, 2019 Computer Detect and - Accuracy
et al., [35] vision and Classify
image Disease
processing
method
Mohan, K. 2016 KNN, SIFT Detect and Problem in Accuracy
J et al., and SVM Classify finding
[36] Disease sickness
Reddy, K. 2017 Image Detect plant More time -
N., [37] processing leaf disease complexity
Methods
Arjunagi S 2017 GLCM Compute the Blighting -

28
et al., [38] SVM amount of Issue
disease
Nidhis, A. 2019 Image Improve the - -
D et al., processing disease
[39] Methods recognition
Kahar, M. 2015 NN Recognise Leaf blast Recognition
A. et al., Fuzzy disease at early disease Rate
[40] Logic stage
Barbedo, 2013 DIP Detect, quantify
Difficult -
J. G. A. et Methods and classify
measure
al., [41] disease symptom of
disease
Anami, B. 2020 Deep-CNN Classify paddy Complexity Accuracy
S et al., crop features
[42]
Rahman, 2020 DNN Automated Complex Accuracy
C. R. et al., process design
[43]
Sharma, R. 2020 CNN Prediction of Infected part -
et al., [44] paddy crop of leaves
features
Das, S. et 2020 Automated Classification Gradient -
al., [45] Process and detection feature
Rautaray, 2020 Automated Detection of Collection Accuracy
S. S et al., Process dataset Issue
[46]

29
 CHAPTER 3

PRESENT WORK

In this present chapter demonstrated the existing problems and studied

various methods of paddy leave disease. The problem statement is defined in section
3.1 and objectives is given as 3.2 and methodology. Generally, the main cause of the
paddy leaves diseases and mainly responsible for damaged crop. Other cause of the
paddy leaves disease of the bacteria light, viral and fungi etc.
3.1 Problem Statement
Plant diseases have turned into a dilemma as it can cause significant
reduction in both quality and quantity of agricultural products. Automatic detection
of plant diseases is an essential research topic as it may prove benefits in monitoring
large fields of crops, and thus automatically detect the symptoms of diseases as soon
as they appear on plant leaves. The proposed system is a software solution for
automatic detection and classification of plant leaf diseases. The scheme consists of
four steps, first a color transformation structure of the input RGB image is created,
and then the green pixels are masked and removed by segmentation process. The
extracted features are passed through the classifier. The disease can be identified and
a solution for disease can be found by the classifier.

The major issues occurred in existing work such as;

1. High time consumption

2. Degrade image quality
3. Less accuracy rates
4. High error rates and
5. Wrong features mostly acceptable.

3.2Research Gaps:
Several approaches for classification approach are used in existing work such
as Deep Neural Network, Artificial Neural Network etc. Various approaches have
used different databases and assumptions to recognize the paddy leaf diseases. S.
Ramesh et al., 2019 [23] classification and identification of Paddy lead diseases

30
using the optimized DNN method has proposed and achieved an accuracy rate of up
to 90 percent. The farmers are fronting various issues such as shortage of water, plant
diseases and natural disasters, etc. Some traditional abilities are overcome the issues
and gaps in the paddy leaf. In proposing method will improve the prevention and
productivity of the plant diseases.

Human visualisation-based method is conservative technique used for the detection

of leaf disease. Generally, the advice of the experts is very costly and consumes
maximum time. Human visualisation technique may suffer from different issues. The
accuracy and precision of human is based on vision of the hired expert. Machine
learning technique is capable to detect the kinds of disease and do right decision and
select the correct treatment. The main advantage of using machine learning method is
more consistent than social experts [39]. Hence, in order overcome the issue of
conventional technique there is requirement of new machine learning (ML) based
classification model. Only little advancement has been done in the field of plant leaf
disease detection using machine learning process and identification and classification
of paddy leaf.
3.3 Research Objectives
The main research work objectives are defined in below:
1. Study the various leaf diseases using recognition and classification methods
in image processing.
2. Develop image pre-processing steps are resize the image, RGB model, HSV
model and clustering method based on image segmentation.
3. Implement a feature extraction and BFO-DNN classification method to
classify the paddy leaf diseases.
4. To compute the performance metrics such as accuracy rate, TPR, F1-Score,
FDR, FPR and compared with the existing parameters.
3.4 Research Methodology
In this section, describe the basic process of the research work based on
classification approach. Firstly, image-processing approaches are developed to the
acquired images to extract the unique features that are important for research work.
The step by step process of research system:

31
1. Search dataset from the online source.
2. Upload leave an image
3. Resize image
4. Convert the RGB input image to HSV format.
5. Image segmentation using K-means Clustering method
6. Extract the unique features (PCA- Principle Component Analysis).
7. Implement BFO-DNN algorithm to classify the paddy leaf diseases.

Input Image

Convert RGB to HSV format

Segmented Image

Extract the Unique Feature Set

Selection of the valuable feature

set

Recognition and Classification

of Paddy leaf diseases

Exit

Fig. 3.1 Main Proposed Flow Chart

32
 8. Compute the performance metrics such as accuracy rate, precision, FDR and
FPR etc.
9. Comparison Analysis.
10. Stop.
In this method develops a new idea to recognize the affective paddy leaf considers to
agricultural field. By the use of clustering method, the disease area of the leaf is
segmented and evaluated. PCA algorithm is used for feature extraction. The feature
extraction method uses to extract the disease paddy leaf, select the feature with the
help of BFOA method and classify the paddy leaf diseases. It has improved the
accuracy rate, reduce the error rate and false classification rate.
Step I: Image Acquisition is a normal process of collecting the database in different
types of paddy leave disease images from the www.PlantVillage.com site which is
used for this research work. The proposed model is using five categories of paddy
leaves like Bacteria Blight, Blast, Brown Spot and etc. In research model is measured
the PL by using HD camera. After that the recognition and classification of disease
all the considered PLIs (Paddy Leave Images) are moved to the computer, where the
development process has been carried out. Sample of PLIs are shown in Fig 15.

Fig.3.2 Input Image

Step II: In this image pre-processing step for reducing the image dimensionality to
rectify the noise in the uploaded images. The important work to remove the image
background by applying the hue –values depends on combined. The PLI in RGB
model is converted into HSV Model. After the noise identification developed a
filtration method such as 3Dbox Filter and Gaussian Filter method to remove the
noise in the uploaded image shown in fig 16.

33
 Fig. 3.3 Filtered Image
Step III: Image Segmentation of PLI using K-means Clustering method is worked in
this research work. The K-means clustering is a main process to divide the classes in
the form of image into clusters. The disease of paddy leave image area is extracted
from the image by using clustering shown in fig 17.

Fig. 3.4 Segmented Paddy Leave Image

Step IV: In this phase has used Feature Extraction using PCA (Principal Component
Analysis) algorithm. It extracts the features in the form of Eigen Values and Eigen
Vectors. It selects the feature using BFOA (Bacteria Foraging Optimization
Algorithm) with Cost function.

(i)

34
 (ii)
Fig. 3.5 (i) Feature Extraction and (ii) Feature Selection.
Step V: The classification procedure is the most vital factor of the PLI disease. The
research model has implemented the DNN (Deep Neural Network) approach to
classify the leave. After that it evaluates the performance metrics such as TPR, TNR,
Accuracy Rate, MSE, Cross Entropy Loss and compared with existing methods
(DNN_JAYA and DNN).

35
 CHAPTER 4

RESULT AND DISCUSIONS

The result analysis chapter contains the mathematical metrics such as

accuracy rate, mse, TPR and TNR etc. It includes the various images working to
compute the classification DNN model for improving the performance analysis and
compared with existing methods.
4.1 Performance Parameters
Accuracy Rate: It is the nearest value to the true value. It is acquired by
minimum readings. The medium reading decreases the error of the calculation.
B8 #= '/""
Accuracy =C=#/'C x100 …………………………. (i)
"# D '/""

TPR (True Positive Rate): It behaves different due to observed fault, It is used for
searching the reliability of the measurement.
CB
Precision =CBECF…………………………. (ii)

TNR (True Negative Rate): It is the measurement of the test accuracy. It is defined
as the weight of harmonic mean of the recall and precision of test.
It is computed formula as:
B8 "= . G /''
F1-Score = …………………………. (iii)
B8 " = EG /''

FPR (False Positive Rate): It is the possibility of false rejection of null hypoProject
for specific testing.
1B
FPR= 1BECF …………………………. (iv)

In equation (iv), FP is false positive, TN is true negative.

FDR(False Discovery Rate): It is the arithmetical method used in diverse

hypoProject testing to correct for numerous comparisons. It is estimated possibility
of false discoveries.
H
FDR = E ( G Where R >0) …………………………. (v)

Here in equation (v), v is amount of type 1 error and r is amount of rejected

hypoProject.

36
Entropy Loss: It is also called as the log loss, considered the performance of the
classification method in which the outcome is the possible value that ranges from 0
and 1. Cross entropy loss increased the possible vale diverge from real label.

It is computed as S and P where;

H (S,P) = - Sum y in Y P (y) * log (Q(y))...............(vi)
4.2 Dataset Used in Paddy Leave Images
Normally, the database was created by managing the manual disease leaves
into several disease categories. They refer the farmer and ask to present the category
of the disease of sample paddy leaves. Farmers have presented in their local
language, recognition and classification of those diseases by referring to agriculture
specialists. This data set has used the recognition and classification of the PLIs.
Image features are extracted, DIP (Digital Image Processing) methods defined the
required input images, image feature extracts and classify the disease.

(a) (b) (c)

Fig. 4.1 Paddy Leaf Dataset Images (a) Bacterial Blight (b) Brown Spot (c) Blast

4.3 Experimental Analysis

The proposed system is simulated in MATLAB used, and GUI has designed a
desktop project application. In this analysis result, totally with 494 images which
includes 102 bacteria blight, 125 brown spot, 86 blast, 61 sheath rot, and 120 normal
are given and image shown in below fig 20.

37
 Fig. 4.2 Sample Paddy Images

Fig. 4.3 Main Project Desktop Application Window

Fig 21 defines the main project desktop application window. The main window
created by MATLAB GUI environment toolbox used. This Project Desktop
Application Window shows the design page and write a code in editor window.
Initially, different types of categories are divided into recognition and classification
system. The proposed system defines the two steps:

(i) Training Model and

(ii) Testing Model

Training Model defines the upload the dataset images one at a time. Image pre-
processing steps to verifying the interference in the upload image. In case,
interference is available then developing the 3D Box and Gaussian Filtration method
to remove the un-wanted noise. HSV Color Space Model is defined that the color-

38
space checked. After that the clustering method to divide image data into various
number of clusters. Image Segmentation phase is find the edge or region of the filter
Paddy Image Leave. Then, it developed the feature extraction methods using PCA.
This approach is used to calculate the unique properties of the image and train the
DNN classification model with dataset and targets.

Testing Model defines the upload the test paddy leave image. It converts the color
image to grey scale image. It segments the region and extracts unique features,
selects the feature set and calculate the parameters.

Fig. 4.4 Upload Paddy Leave test image, Gray image, HSV image, Noise image, and
Filter Image.

Above fig 22 defines the PIL disease category image is uploaded test image. It shows
the conversion the 3dimensional into 2dimensional image means output is grey
image. It identifies the noise data in the paddy image. Then, applied the different
types of filtration methods such as:

(i) 3D Box Filtration and

(ii) Gaussian Filter.

These filtration methods are applied to remove the noise in the converted image. It
implemented the K-means clustering methods to calculate the regions of the paddy
leave disease image. It extracts the defected area or region in the filtered image.

39
 Fig. 4.5 Feature Extraction and Selection

The above fig 23. shows the feature extraction graph using PCA method. This
method is extracting the feature values and extract the region values. It finds the local
kind of feature values and extract the region points. In this feature selection
optimization method to selects the valuable extracted features and shown in graphical
format. This method is based on performance metric modification, varied search-
strategy and change the solution space to create the search simply using various
reproductions.

Fig. 4.6 Category recognize and classification Image

Above fig 24 shows the paddy leave disease recognition and classification in the
images. The recognition and classification procedure is done by BFOADNN
algorithm.

40
 Fig. 4.7 Accuracy with Proposed Method Using Hybrid BFO- DNN.
Above fig 25 shows the research model performance with the accuracy rate
evaluated by BFOA-DNN classification model. The proposed method has
implemented in the paddy leave disease recognition and classification system to
increase the accuracy parameter as compared with the existing DNN-JAYA and
DNN method. The proposed system accuracy has increased, according to the
increase the number of samples. The research system accuracy rate value is 98%.

Fig. 4.8 False Discovery Rate with Proposed Method Using Hybrid BFO- DNN.

41
The above Fig. 26 shows the FDR performance is the predictable proportion of kind
of exceptions. The main kind of exception is where incorrectly discard the null
hypoProject. Thus, FDR value is 0.0011.
Cross Entropy Loss
10 -3
Hybrid BFO-DNN

10 -4

10 -5

10 -6
0 10 20 30 40 50 60 70 80 90 100
Number of samples

Fig. 4.9 Cross Entropy Loss with Proposed Method Using Hybrid BFO- DNN.

Above fig 27 shows the research model performance with the Cross Entropy Loss
evaluated by BFOA and DNN classification model.The research model has
implemented in the PLI disease recognition and classification system to decrease the
entropy loss parameter as compared with the existing algorithms. The proposed
model entropy loss has decreased, according to the increase the probability. The
proposed entropy lossvalue is 0.00020965%.

True Positive Rate

10 0
Hybrid BFO-DNN

10 -1

10 -2
0 10 20 30 40 50 60 70 80 90 100
Number of samples

Fig. 4.10 True Positive Rate with Proposed Method Using Hybrid BFO- DNN.

42
Above fig 28 shows the deep learning method using BFOA-DNN algorithm, the TPR
is also denoted to recall, is utilized to consider the percentage of real positives which
are appropriately verified. Thus, the true positive rate (TPR) is 0.9988.

Fig. 4.11 True Negative Rate with Proposed Method Using Hybrid BFO- DNN.

Above fig 29 shows the deep learning using DNN method, the TNR is also denoted
to specifically, considers the proportion of real incorrectly values that are correctly
verified. Thus, the true negative rate value is 0.991%.

Fig. 4.12 Comparison Analysis with Accuracy Rate (%)

43
 Fig. 4.13 Comparative Analysis with Entropy Loss (%)

Above fig 30 and 31 shows the comparative analysis with accuracy rate and Entropy
Loss parameters. The research method has improved the accuracy rate as compared
with different type of deep learning methods such as:
(i) DNN (Deep Neural Network).
(ii) Deep Neural Network with Jaya Optimization, and
(iii) Deep Neural Network with Bacterial Foraging Optimization.

Table 4.1 Proposed Parameter with Hybrid BFOA-DNN method

Parameters Values

Cross Entropy Loss (%) 0.000209

Accuracy Rate (%) 98

TNR (%) 0.991

TPR(%) 0.998

FPR (%) 0.0098

FDR (%) 0.0011

44
Table 2 shows the proposed Hybrid BFOA-DNN model work performance with
parameters such as accuracy rate value is 98 %, Cross Entropy Loss value is 0.0002,
TNR, TPR values are 0.991, 0.9984, FPR value is 0.0098 and FDR value is 0.0011.
In proposed model has increased the accuracy rate and reduced the cross-entropy loss
in the paddy leave disease classification and recognition system.
Table 4.2 Existing Performance
Parameters DNN_JAO DNN Classifier

Cross Entropy Loss (%) 0.0100 0.01700

Accuracy Rate (%) 97 93.50

Above table 3 shows the system performance with existing methods using DNN with
JAO algorithm accuracy rate is 97% and DNN classifier accuracy performance is
93.50%. The existing method cross entropy loss performance value is 0.010 and
DNN classifier value is 0.01700.
Table 4.3 Comparison: Proposed and Existing Models (Hybrid, DNN+JAO, DNN)
Parameters Hybrid BFOA- DNN_JAO DNN
DNN

Accuracy Rate 98 97 93.50

(%)

Cross Entropy 0.000209 0.0100 0.01700

Loss (%)

Table 4 shows the comparative analysis with proposed and existing methods such as
hybrid BFOA-DNN, DNN and DNN with Jaya optimization method. The proposed
system performance value of accuracy is 98% with hybrid BFOA-DNN, accuracy
value is 97 and DNN 93.50 percentage. The research system performance value of
Cross Entropy Loss is 0.0011 with hybrid BFOA-DNN, Entropy Loss value is
0.0100 and DNN 0.01700 per cent.

45
 CHAPTER 5

CONCLUSION AND FUTURE SCOPE

5.1 Conclusion
In agriculture, the loss of yield mainly takes place due to the increasing rate
of the disease. Generally, the detection and classification of the disease is determined
when it reaches at the harsh stage. Hence, there is loss of the yield, time and wealth.
The planned method is capable of the detection of the disease at former stage as soon
as it takes place on the leaf. However, the reducing the reliance and saving the loss
rate on specialist to definite amount is promising. It is provided with the help of the
individual having minimum information of the disease. The detection of the plant
leaves disease through some automated method is valuable as it decreases the
monitoring load of the crops and identifies the symptom at previous stage.

In proposed work, the classification and identification of various leaf diseases using
image processing method. The paddy leaves are captured directly from the farm field
for standard and diseases such as bacterial blight, brown spot, sheath spot and blast
images. During the pre-processing, the contextual data are removed, RGB pictures
are transformed into HSB pictures. In the initial stage, the different paddy leaves
acquired through digital images. Then, the RGB model transformed into HSV model
toresizethe image using k mean clustering with image segmentation method. And,
then the unique features are extracted using PCA algorithm. In addition,
implemented feature extraction and BFO-DNN method for the classification of the
paddy leaf diseases. Numerical analysis is done to compute the performance metrics
like as accuracy, entropy loss,FDR and FPR. The simulation tool is used in
MATLAB in proposed system. It compared with software reading and base paper
reading the accuracy 1% improved. GUI is developed a desktop project application
with layout editor. The main advantage of the research software system is that it can
be simply extended for other leaves with minor changes. The accuracy performance
has been enhanced by using a dataset with HD images.

46
5.2 Future Scope

Future scope will attempt to develop the method that identify and detect the
plant leaf diseases. It will be beneficial for the farmers to diagnose the leaves
infection at early stage. In addition, image processing web-based method developed
that helped in the classification of the leaf’s diseases.

47
 REFERENCES
[1]Mukherjee, M., Pal, T. and Samanta, D. (2012), “ Damaged paddy leaf detection
using image processing,” Journal of Global Research in Computer Science, 3(10),
07-10.
[2] Kurniawati, N. N., Abdullah, S. N. H. S., Abdullah, S. and Abdullah, S. (2009), “
Investigation on image processing techniques for diagnosing paddy diseases,”
In 2009 international conference of soft computing and pattern recognition (pp. 272-
277). IEEE.
[3]Rathod, A. N., Tanawal, B and Shah, V. (2013), “ Image processing techniques
for detection of leaf disease,” International Journal of Advanced Research in
Computer Science and Software Engineering, 3(11).
[4]Kurniawati, N. N., Abdullah, S. N. H. S., Abdullah, S, “Investigation on image
processing techniques for diagnosing paddy diseases,” In 2009 international
conference of soft computing and pattern recognition (pp. 272-277). IEEE, 2009.
[5]Narmadha, R. P and Arulvadivu, G, “Detection and measurement of paddy leaf
disease symptoms using image processing,” In 2017 International Conference on
Computer Communication and Informatics (ICCCI) (pp. 1-4). IEEE, 2017.
[6]Patel, A., & Joshi, B. (2017), “ A Survey on the Plant Leaf Disease Detection
Techniques,” International Journal of Advanced Research in Computer and
Communication Engineering, 6(1).
[7] Schulz, H., Schrader, B., Quilitzsch, R., Pfeffer, S. and Krüger, H. (2003), “
Rapid classification of basil chemotypes by various vibrational spectroscopy
methods,” Journal of agricultural and food chemistry, 51(9), 2475-2481.
[8] Cubadda, F., Aureli, F., Ciardullo, S., D’Amato, M., Raggi, A., Acharya, R and
Prakash, N. T. (2010), “ Changes in selenium speciation associated with increasing
tissue concentrations of selenium in wheat grain,” Journal of agricultural and food
chemistry, 58(4), 2295-2301.
[9] Ku, W., Storer, R. Hand Georgakis, C. (1995), “ Disturbance detection and
isolation by dynamic principal component analysis,” Chemometrics and intelligent
laboratory systems, 30(1), 179-196.

48
[10] Kim, H. W., Kim, T. K., Kee, S. C., Lee, J. H and Hwang, W. J. (2007). U.S.
Patent No. 7,254,257. Washington, DC: U.S. Patent and Trademark Office.

[11]Li, K., Liu, Y., Wang, Q., Wu, Y., Song, S., Sun, Y. and Du, S. (2015), “ A
spacecraft electrical characteristics multi-label classification method based on off-
line FCM clustering and on-line WPSVM,” PloS one, 10(11).
[12]Sharma, V., Pattnaik, S. S., Garg, T, “A review of bacterial foraging
optimization and its applications,”International Journal of Computer Applications
(IJCA), 2012.
[13] Wu, X. L., Zhang, Z. Q., Du, Y. H., YAN, J, “Improved bacteria foraging
optimization algorithm for flexible job shop scheduling problem,” Computer
Integrated Manufacturing Systems, vol. 21(5),pp 1262-1270, 2015.
[14] Afandie, W. N. E. A. W., Rahman, T. K. A. and Zakaria, Z. (2016), “
Comparative Analysis of Bacterial Foraging Optimization Algorithm and
Evolutionary Programming for Load Shedding in Power System,” International
Journal of Simulation--Systems, Science & Technology, 17(41).
[15] Montavon, G., Samek, W., Müller, K. R, “Methods for interpreting and
understanding deep neural networks,” Digital Signal Processing, 73, pp 1-15, 2018.
[16] Deng, L., Hinton, G., & Kingsbury, B. (2013), “ New types of deep neural
network learning for speech recognition and related applications: An overview,”
In 2013 IEEE International Conference on Acoustics, Speech and Signal
Processing (pp. 8599-8603). IEEE.
[17]Bahi, M and Batouche, M. (2018), “ Deep Learning for Ligand-Based Virtual
Screening in Drug Discovery,” In 2018 3rd International Conference on Pattern
Analysis and Intelligent Systems (PAIS) (pp. 1-5). IEEE.
[18] Srimani, P. K and Nithiyanandhan, K. (2016), “ Analysis of the Leaf Histogram
with HSV-Model,” International Journal of Engineering and Management Research
(IJEMR), 6(2), 64-68.
[19]Zheng, X., Lei, Q., Yao, R., Gong, Y., Yin, Q. , “Image segmentation based on
adaptive K-means algorithm,” EURASIP Journal on Image and Video
Processing, 2018(1), 68, 2018.

49
[20]Kumar, D., Kumar, S and Rai, C. S. (2009), “ Feature selection for face
recognition: a memetic algorithmic approach,” Journal of Zhejiang University-
Science A, 10(8), 1140-1152
[21] Hernández-Ocana, B., Mezura-Montes, E. and Pozos-Parra, P. (2013), “ A
review of the bacterial foraging algorithm in constrained numerical optimization,”
In 2013 IEEE Congress on Evolutionary Computation (pp. 2695-2702). IEEE.
[22] Panda, M. (2017), “ Elephant search optimization combined with deep neural
network for microarray data analysis.,” Journal of King Saud University-Computer
and Information Sciences.
[23] Ramesh, S., and Vydeki, D. “Recognition and classification of paddy leaf
diseases using Optimized Deep Neural network with Jaya algorithm”, Information
Processing in Agriculture,(2019).
[24] Prajapati, H. B., Shah, J. P., and Dabhi, V. K. “Detection and classification of
rice plant diseases”, Intelligent Decision Technologies, 11(3),(2017), 357-373.
[25] Ramesh, S. “Rice Blast Disease Detection and Classification Using Machine
Learning Algorithm”. In 2018 2nd International Conference on Micro-Electronics
and Telecommunication Engineering (ICMETE) (pp. 255-259),(2018). IEEE.
[26] Shah, J. P., Prajapati, H. B., and Dabhi, V. K. “A survey on detection and
classification of rice plant diseases”,In 2016 IEEE International Conference on
Current Trends in Advanced Computing (ICCTAC) (pp. 1-8),(2016). IEEE.
[27] Narmadha, R. P., and Arulvadivu, G. “Detection and measurement of paddy
leaf disease symptoms using image processing”,In 2017 International Conference on
Computer Communication and Informatics (ICCCI) (pp. 1-4),(2017). IEEE.
[28] Islam, T., Sah, M., Baral, S., and RoyChoudhury, R. “A Faster Technique on
Rice Disease Detectionusing Image Processing of Affected Area in Agro-
Field”,In 2018 Second International Conference on Inventive Communication and
Computational Technologies (ICICCT) (pp. 62-66), (2018). IEEE
[29] Singh, A., and Singh, M. L. “Automated color prediction of paddy crop leaf
using image processing”,In 2015 IEEE Technological Innovation in ICT for
Agriculture and Rural Development (TIAR) (pp. 24-32),(2015). IEEE.

50
[30] Singh, A., and Singh, M. L. “Automated blast disease detection from paddy
plant leaf—A color slicing approach”,In 2018 7th International Conference on
Industrial Technology and Management (ICITM) (pp. 339-344),(2018),IEEE.
[31] Zhou, L., and Li, C. “Outsourcing eigen-decomposition and singular value
decomposition of large matrix to a public cloud”, IEEE Access, 4,(2016), 869-879.
[32] Pinki, F. T., Khatun, N., and Islam, S. M. “Content based paddy leaf disease
recognition and remedy prediction using support vector machine”,In 2017 20th
International Conference of Computer and Information Technology (ICCIT) (pp. 1-
5),(2017), IEEE.
[33] Suresha, M., Shreekanth, K. N., and Thirumalesh, B. V. “Recognition of
diseases in paddy leaves using knn classifier”,In 2017 2nd International Conference
for Convergence in Technology (I2CT) (pp. 663-666),(2017). IEEE.
[34] Gu, X., Zhang, J., Yang, G., Song, X., Zhao, J., and Cui, B. “Inversion of paddy
leaf area index using Beer-Lambert law and HJ-1/2 CCD image”,In 2013 IEEE
International Geoscience and Remote Sensing Symposium-IGARSS (pp. 2794-
2797),(2013), IEEE.
[35] Kapilya G, G. Rosline Nesa Kumari, D. Dhanasekaran (2019), “Classification
and Functional Analysis of Major Plant Disease using Various Classifiers in Leaf
Images”, International Journal of Innovative Technology and Exploring Engineering
(IJITEE).

51
 LIST OF PUBLICATIONS

1. Plant Disease Detection by Image Processing – A Review,International Journal of

Scientific Research in Engineering and Management, July 2020 edition.

2. “Paddy Leaf Diseases Recognition and Classification using PCA and BFO-DNN
Algorithm by Image Processing"- International Conference on Advances in
Communications, Computing and Embedded Systems (ACCES-2020), Elsevier
scopus indexed proceedings
(Accepted).https://www.journals.elsevier.com/materials-today-proceedings

3. Abcd, additional paper information

52
Plagiarism Report

53