ABSTRACT

Traditional farming is going out of date nowadays. Technologies are being

introduced in the farming sector for the past decade and in recent years it is seen
that the participation of deep learning and machine learning is playing an integral
role in solving traditional problems. The introduction of new technology has
increased the productivity of farmers and also increased the yields and quality of the
crops too. Plant diseases are a serious concern for the consumers and the farmers
too. It does not only carry some harmful bacteria within itself however it
compromises the yield of the crops too. The identification of such plant diseases has
been a continuous problem for cultivators and researchers. Deep learning-enabled
developments in the field of computer vision have paved the path for computer-
assisted plant disease diagnosis. Deep Learning has achieved great success in the
categorization of a number of plant diseases by exploiting its ability to recognize
objects with the help of convolutional neural networks. Various deep learning
algorithm like AlexNet and LeNet-5 is applied on a publicly available dataset
(plantvillage dataset) so that the neural network can capture the various features of a
specific disease and diagnose it accordingly using a human-like decision making skill
 TABLE OF CONTENT

SL.NO TITLE PAGE.NO

01. INTRODUCTION 1
1.1 OUTLINE OF THE PROJECT 1
1.2 OBJECTIVES 2

02. LITERATURE SURVEY 3-7

03. AIM AND SCOPE 7

3.1 PROJECT GOAL 7
3.2 SCOPE OF THE PROJECT 8
3.3 OVERVIEW OF THE SYSTEM 9-11

04. METHODOLOGY 12
4.1 SYSTEM ARCHITECTURE 18-23
4.2 TYPES OF CNN 23-27

05. COMPARISION AND ANALYSIS 27-29

06. SUMMARY, CONCLUSION AND FUTURE WORK 30-32

07. APPENDICES 32
7.1 SOURCE CODE 32-55
56-59
7.2 SCREENSHOTS
60-61
7.3 PUBLICATION AND PLAGIARISM REPORT 62-63
7.4 REFERENCE

LIST OF FIGURES
SL.NO TITLE PAGE.NO
1. DATAFLOW DIAGRAM 2
2. SYSTEM ARCHITECTURE 18
3. WORKFLOW DIAGRAM 19
4. USECASE DIAGRAM 19
5. CLASS DIAGRAM 20
6. ACTIVITY DIAGRAM 21
7. SEQUENCE DIAGRAM 21
8. E.R – DIAGRAM 22
9. COLLABORATION DIAGRAM 23

LIST OF SYSMBOLS

NOTATION
S.NO NAME NOTATION DESCRIPTION

Class Name
1. Class Represents a
+ public collection of similar
-attribute entities grouped
-private
-attribute together.
# protected

+operation

+operation

+operation

NAME

Class A Class B

Associations
represents static
2. Association
Class A Class B relationships between
classes. Roles
representsthe way the
two classes see each
other.

3. Actor It aggregates several

 classes into a single
classes.

Class A Class A
Interaction between
Aggregation the system and
4.
external environment
Class B Class B

5. Relation uses Used for additional

(uses) process
communication.

Extends relationship is
used when one use
6. Relation EXTENDS
case is similar to
(extends) another use case but
does a bit more.

7. Communication Communication
between various use
cases.
8. State State State of the processs.

9. Initial State Initial state of the

object

10. Final state

F inal state of the

object

11. Control flow

Represents various
control flow between
the states.

Represents decision
making process from a
12. Decision box
constraint

13. Interact ion between

Usescase the system and
Usecase
 external environment.

Represents physical
modules which is a
14. Component
collection of
components.

Represents physical
modules which are a
15. Node
collection of
components.

A circle in DFD
represents a state or
16. Data
process which has
Process/State
been triggered due to
some event or acion.

17. External Represents external

entity entities such as
keyboard,sensors,etc.
18. Transition Represents
communication that
occurs between
processes.

19. Object Represents the vertical

Lifeline dimensions that the
object
communications.

20. Message Message Represents the

message exchanged.
1. INTRODUCTION

India being an agriculture country, about 70% of the population depends on it as

their main source of income and food. Agriculture plays and important part of the
Indian economy as it contributes about 17% of the total GDP. Farmers have wide
range in selecting their crops and finding a suitable pesticide for it but in spite of
all their efforts it can all be vain if they can‟t identify the disease plaguing their
crops. Thus, disease on crops can significantly reduce the quality and quantity of
agricultural products along with economical damage to the farmers. To
successfully cultivate crops without incurring much loss we need to properly
identify the disease and remedy it, this requires a lot of work and processing time
as detecting each and every plant can be tedious can time consuming. To lessen
the burden of the farmers along with their losses we propose the use of a system
which can detect infected plants so that we can curb the spread of infection and
diseases at an earlier step thus reducing losses and crop failure.

In most cases symptoms like fungal infection and rot can be seen on the leaves,
stem and fruit. This project provides an insight into how we deal with the problem
and further discuss the challenges of our work and how we can improve upon it
in future work.

1.1 OUTLINE OF THE PROJECT

Overview of the system:

 Define a problem
 Gathering image data set
 Evaluating algorithms
 Detecting results

The steps involved in Building the data model is depicted below.

1
 Data collection (Splitting Training set & Test)
set)

Pre Processing (Sequential)

Building classification Model

Prediction (Plant leaf disease Prediction)

Fig 1: Data flow diagram for CNN model

1.2 OBJECTIVE :

Smart farming system using necessary infrastructure is an innovative technology

that helps to improve the quality and quantity of agricultural production in the
country. Disease in plants has long been one of the major threats to food
security as it dramatically reduces the crop yield and compromises the quality.
The identification of such diseases has been a significant challenge to cultivators
and researchers. Deep learning-enabled developments in the field of computer
vision have paved the path for computer-assisted plant disease diagnosis. Deep
learning with convolutional neural networks (CNN) has achieved tremendous
success in the categorization of a number of plant diseases by exploiting its
ability to recognise objects, and the solution provides an efficient technique for
detecting plant disease. Various CNN algorithm like AlexNet and LeNet-5 is
applied on a publicly available dataset (plant village dataset) so that the neural
network can capture the various features of specific disease and diagnose it
accordingly using a human-like decision making skill.

2
 The presented work presents a color based segmentation techniques for
extraction of yellow rust in whet crop images. Accurate segmentation of yellow rust in
wheat crop images is very part of assessment of disease penetration into the wheat
crop. And in turn to take the necessary preventive action for minimizing the crop
damage. The jpeg images acquired from CCD camera are read into the matlab tool
and a color-based segmentation algorithm is performed to segment the yellow rust.
The segmentation of color is performed base on k-means algorithm.

TITLE: Comparative study of Leaf Disease Diagnosis system using Texture features
and Deep Learning Features

AUTHOR: Ashwini T Sapka, Uday V Kulkarni

YEAR: 2018

The feature extraction technique plays a very critical and crucial role in
automatic leaf disease diagnosis system. Many different feature extraction
techniques are used by the researchers for leaf disease diagnosis which includes
colour, shape, texture, HOG, SURF and SIFT features. Recently Deep Learning is
giving very promising results in the field of computer vision. In this manuscript, two
feature extraction techniques are discussed and compared. In first approach, the
Gray Level Covariance Matrix(GLCM) is used which extracts 12 texture features for
diagnosis purpose. In second appraoch, the pretrained deep learning model, Alexnet
is used for feature extraction purpose. There are 1000 features extracted
automatically with the help of this pretrained model. Here Backpropagation neural
network (BPNN) is used for the classification purpose. It is observed that the deep
learning features are more dominant as compared to the texture features. It gives
93.85% accuracy which is much better than the texture feature extraction technique
used here.