Professional Documents
Culture Documents
by
Rakhunde Pranav (B190358564) Sawant Sourabh(B190358566)
Yadav Sanket (B190358580) Sawant Yash(B190358581)
Certificate
This is to certify that the project report on
Citrus Plant Disease Classifcation and
Recommendation System Using Deep Learning
submitted by
Rakhunde Pranav (B190358564) Sawant Sourabh (B190358566)
Yadav Sanket(B190358580) Sawant Yash(B190358581)
in the completely fulfillment of the requirement for the award of
Degree of Bachelor of Engineering in Information Technology at Vidya
Pratishthan’s Kamalnayan Bajaj Institute Of Engineering and
Technology, Baramati under the Savitribai Phule Pune University,
Pune. This work is done during year 2022-23 Sem-II, under our guidance.
i
Acknowledgements
I feel happy in forwarding this seminar report as an image of sincere eort. The successful
project reects my work, eort of my guide in giving me good information. I express my
deep gratitude to all sta members who lend me their valuable support and cooperation
to enable me to complete my seminar successfully. My sincere thanks to my guide
Mr. B. S. Salve sir who has been constant source of inspiration and guiding star in
achieving my goal. I give my special thanks to respected Prof. P.M.Patil sir who is
Project coordinator and Dr. S.A.Takale ,Head of Information Technology department
for their constant interest and encouragement throughout the completion of our project.
I am also equally indebted to our principal Dr.R.S.Bichkar sir for his valuable help
whenever needed.
Thank You!
Pranav Rakhunde
Sourabh Sawant
Sanket Yadav
Yash Sawant
1
Abstract
Nowadays, with the development in machine learning and deep learning, it is possible to
not only identify but control plant diseases by using computer devices instead of manual
inspection. So it helped in improving the quality as well as production of fruits. Citrus
fruits are well known for their taste and nutritional value. They are one of the natural
and well-known sources of vitamin C, so planted worldwide. Manual inspection of fruit
diseases with the naked eye takes time and also it is dicult. Therefore, a computer-
based method is always required for the accurate recognition of plant diseases. So build
a citrus image dataset including all common citrus diseases.
The deep learning network is used to train and learn these images, Which can ef-
fectively identify and classify plant diseases. The CNN model is used as the primary
network. The results show that techniques of data augmentation and preprocessing have
delivered promising insights to estimate citrus fruit’s damages in percentage and also
identify and classify agricultural diseases with relevant suggestions through which can
avoid citrus plant disease with specic insecticides, Pesticides, and fertilizers.
2
Contents
Acknowledgements 1
Abstract 2
Keywords 7
1 Introduction 9
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Literature Survey 11
3 Proposed Work 15
3.1 Problem Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Scope of Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Project Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.4 Project Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5 Systems Architecture 19
6 Project Planning 21
6.1 Project Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.2 Team Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
7 Project Schedule 25
7.1 Project Breakdown Structure . . . . . . . . . . . . . . . . . . . . . . . . 25
7.2 Task Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3
CONTENTS CONTENTS
8 Project Design 28
8.1 DataFlow Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.1.1 Dataow Level 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.1.2 Dataow Level 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.1.3 Dataow Level 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
8.2 UML Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.3 Usecase Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.4 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8.5 Class Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
10 Test Cases 40
11 Experimental Results 42
11.1 GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
11.1.1 Load Image Module . . . . . . . . . . . . . . . . . . . . . . . . . . 42
11.1.2 Process Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
11.1.3 Output Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
11.2 Working Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
11.2.1 Disease Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
11.2.2 Recommendation based on diseases . . . . . . . . . . . . . . . . . 45
11.3 Experimental Results and Discussions . . . . . . . . . . . . . . . . . . . . 46
11.3.1 Confusion Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
11.3.2 Precision, Recall and F1 Score . . . . . . . . . . . . . . . . . . . . 46
11.3.3 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
11.3.4 Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
References 50
A Appendix 51
B Base Paper 52
C Tools Used 53
D Papers Published/Certicates 54
4
List of Figures
5
List of Figures
6
Keywords
List of keywords -
• Deep Learning
• Image Classication
• Feature Selection
• Data Augumentation
7
Notation and Abbreviations
List of Notations -
8
Chapter 1
Introduction
1.1 Introduction
There are many types of agricultural calamities with great impact and frequent disasters,
which not only cause losses to crop production but also threaten food safety. Crop disease
is one of the main types of “disasters” which aect the citrus plant. Citrus disease is one
of the most important trading crops all over the world. It brings great economic losses to
farmers. There are some diseases that are such as Greening, Canker, Scab, Black spot,
and Melanose. It takes selects several common citrus diseases as experimental objects
disease problems certainly aect crops throughout their growing cycle.
In the agriculture eld, disease detection is an important step. If the disease is not
detected at an early stage will lead to farmers’ economic loss. Machine vision framework
uses image processing techniques to perform such specic work, which is why image pro-
cessing assumes an exceptionally signicant job in their capabilities. The system will
take input as images of citrus plant and gives output as detected diseases for which plant
is infected and provide the recommendation by proving required measurements of fertil-
izers or pesticides through CNN and classication algorithms. So the system is proposed
to classify the diseases using Convolutional Neural Network.
9
1.2. MOTIVATION CHAPTER 1. INTRODUCTION
1.2 Motivation
The main motivation behind this project is that it Proposed CNN-based citrus plant
disease identication model is capable of distinguishing between healthy and diseased
citrus fruits and leaves. We used the CNN model to tackle the problem of classifying
diseases from citrus fruit and leaf images in this study. The modules in our proposed
model are as follows: i) Data loading, ii) Data preprocessing, and iii) building CNN
model. Collects high-level attributes, yielding disease classication of citrus fruit/leaves
into Black spot, Canker, Scab, Greening which not the human eye can’t see. It also not
only detects the disease but also predicts the required fertilizers by which it prevents the
disease the increase the citrus yielding.
Literature Survey
This system used computer vision as well as deep learning algorithms such as Convolu-
tional Neural Network, Pretrained Mobile Net-V2, Whale Optimization for optimization
of the features. Firstly, data of images citrus was augmented. Later on, It was trained
by using pre-trained network Mobile Net V2 CNN and ne-tuned. The newly trained
model was used for deep feature extraction. After that used Whale Optimization algo-
rithm (WOA) to remove redundant information present in the features. Lastly, images
were classied into dierent kinds of diseases.
Technique Used:
• Mobile Net V2
• Data Augmentation
• CNN
11
CHAPTER 2. LITERATURE SURVEY
In this system Deep learning algorithms have been implemented for citrus disease
detection to classify fruit into healthy and defective classes. First, a dense CNN model
was used without performing preprocessing and data augmentation on 150 frames and
achieved an accuracy of 67 percent, But the proposed model used data augmentation
and preprocessing to improve CNN performance and used 1200 frames. Furthermore,
the proposed model is compared with a dense model where data augmentation and
preprocessing techniques were not used. The overall accuracy of the proposed model is
89.1 percent. The results show that the data augmentation and preprocessing techniques
yielded promising insights for damage estimation of citrus fruits.
Technique Used:
• Data Augmentation
• Image Processing
3. Wenyan Pan, Jiaohua Qin,Xuyu Xiang, Yan Wu1, Yun Tan, and Lingyun
Xiang, ”A Smart Mobile Diagnosis System for Citrus Diseases Based on
Densely Connected Convolutional Networks”, date of publication June 26,
2019; date of current version July 18, 2019. Digital Object Identier 10.1109/AC-
CESS.2019.2924973
In this system, they had taken image dataset of six types of citrus diseases and made a
intelligent diagnostic system for citrus diseases by creating a simplied densely networked
convolutional networks (DenseNet). The system is implemented using the WeChat applet
on a mobile device, through which users can upload images and receive diagnostic results
and comments. Experimental results showed that citrus disease recognition accuracy
exceeds 88 percent. Also, prediction time had decreased due to use of DenseNet structure.
Technique Used:
• Dense-Net
• Intelligent diagnosis
• WeChat applet
In this system, a new deep learning-based technique for citrus disease classication is
proposed. In this work,two dierent pre-trained deep learning models were used. Image
scaling techniques are used to increase the size of the citrus dataset used in this paper. In
addition, hybrid contrast stretching was used to improve the visual quality of the images.
In addition, transfer learning is used to retrain pre-trained models, and the feature set is
enriched with the use of fusion. The merged feature set is optimized using a metaheuristic
algorithm, the Whale Optimization Algorithm (WOA). The selected characters are used
to classify six dierent diseases of citrus plants. The proposed technique achieves a
classication accuracy of 95.7 percent with excellent results compared to state-of-the-art
techniques.
Technique Used:
• Mobile Net V2
• CNN
• Data Augmentation
• Dense Net
• Transfer Learning
Proposed Work
In agriculture eld, disease detection is important step. If disease is not detected at early
stage will lead to farmer’s economical loss. The system will take input as images of citrus
plant and gives output as detected diseases through CNN and classication algorithms.
So system is proposed to classify the diseases and recommend the pesticides.
• The proposed system is able to detect the dierent citrus plant diseases such as
Black spot, canker, scab, greening, and Melanose.
• Citrus plant disease classication has practical applications in the eld of plant
disease recognition based on their visual symptoms.
• The system will be able to detect diseases which are at early stage of the disease
and recommend the fertilizers according to the disease.
15
3.3. PROJECT OBJECTIVES CHAPTER 3. PROPOSED WORK
• It detects only disease but it is unable to give the exact damage i.e. how much plant
is infected with the disease.
• We only used one deep learning-based CNN model in this system other deep learning
models can be also used.
For the best performance of the software user must follow the sequence of the activities
to achieve the required results. Do not proceed to recognise text before the picture is
captured. While using the software, users action must be consistent and unique. input
to the software must be in proper format.
• Flexibility : The architecture of the application will be exible enough for some
later requirements change or application enhancement.
17
4.3. SECURITY REQUIREMENTS CHAPTER 4. PROJECT REQUIREMENT SPECIFICATION
Security requirements are needed to prevent any malicious attack that can take place
on the project. These requirements are as follows:-
• The type of data is produced at the website should not be exposed to fertilizers
company, so that they could not manipulate the market.
• RAM : Minimum 8 GB
Systems Architecture
19
CHAPTER 5. SYSTEMS ARCHITECTURE
Firstly system will take input images of the citrus plant which healthy as well as af-
fected by the diseases. After that it will be given to the next step and data is downloaded
from the keras. After that data is splitted into training data, testing data and validated
data is preprocessed i.e. resized and rescaled. Also data is augmented. Input layers are
dened and hidden also dened. As output needed for data is data classied into ve
dierent classes.
So, ve neurons are passed at the output layer. Here, CNN is used which contains
ve convolutional with simultaneous max-pooling layers. Model is build with the data
as well as performance also evaluated. Dierent images were tested with random data
of citrus plant and output is checked. Lastly based on disease fertilizer is recommended
with the help of fertilizer dataset.
Project Planning
21
6.1. PROJECT ESTIMATES CHAPTER 6. PROJECT PLANNING
Budget Estimation:
Estimation is the process of nding an estimate, or approximation, which is a value that
can be used for some purpose even if input data may be incomplete, uncertain, unstable.
Estimation determines how much money, eort, resources, and time it will take to build
a specic system or product Cost of Project:
Equation for calculation of cost of project using COCOMO model is:
C = D * Cp
Where,
C = Cost of project
D = Duration in month
Cp= Cost incurred per person-month,
Cp=Rs.3000/- (per person-month) (approx.)
C = 9 * 3000
= 27000/- Hence according COCOMO model the cost of project is 27000/-(approx)
Project Schedule
25
7.2. TASK NETWORK CHAPTER 7. PROJECT SCHEDULE
Project Design
28
8.1. DATAFLOW LEVELS CHAPTER 8. PROJECT DESIGN
Dataset Description:This Dataset taken from our base research paper.We used dataset of
leaves in our project,which contains 1036 images spread across 5 classes.This dataset con-
tains various types of disease such as canker,black spot,mealnose,healthy,greening.The
rst le contains 171 images of black spot,the second le contains 163 images of canker,the
third contains 204 images of greening,the four contains 485 images of healthy,the last
one contains 13 images of melanose.The Leaves datsets contains images that facilitates
training ,testing and validation during utilization of deep learning algorithun for plant
disease classication or recommendations.
34
9.2. CNN ALGORITHM CHAPTER 9. SYSTEM ALGORITHMS AND IMPLEMENTATION
Step 1 :
Step 2 :
Image Input. - The input layer includes the Convolution layer, ReLu activation layer,
and layer(max) pooling.
Step 3 :
Step 4 :
Step 5 :
Read the image from the training directory and applied image pre-processing techniques-
data cleaning, data transformation, and dimensionality reduction.
Step 6 :
Step 7 :
9.3 Code
Chapter 10
Test Cases
Experimental Results
11.1 GUI
Screenshot 1: Input
This screenshot shows the graphical user interface (GUI) of the Load Image module.
The user is presented with a button labeled ”Upload Image” that allows them to select
an image le from their device. The user can click this button to open a le dialog and
choose the desired image.
42
11.1. GUI CHAPTER 11. EXPERIMENTAL RESULTS
Screenshot 2: Process
In this screenshot, we can see the GUI of the Process module. The uploaded image is
displayed in a designated area. Below the image, there is a button labeled ”Predict”
that triggers the classication of disease. Clicking this button initiates the algorithm to
process the image and classify the disease and recommend fertilizer.
Screenshot 3: Output
The Output module screenshot showcases the result of the disease classication process.
A text box is provided where the disease name is displayed. The user can see the disease
name in this area. Additionally, system recommended the fertilizer according to the
disease.
There are some diseases that are the most dangerous, such as black spot,canker,scab,greening
and melanose classes.It takes selects several common citrus diseases as experimental ob-
jects Disease problems certainty aect crops throughout their growing cycle in agriculture
eld,disease detection is important step.If disease is not detected at early stage will lead
to farmer’s economical loss.Machine vision framework uses image processing techniques
to perform such specic work,that is why image processing assumes an exceptionally
signicant job in their capabilities.
The rst model predicts the disease due to which plant is infected and then by display-
ing not only disease name but also gives the recommendation that how to degrade this
disease and which fertilizers,insecticides and pesticides that used to overcome this prob-
lem.It recommend the user that what they should be done.
11.3.3 Accuracy
11.3.4 Loss
12.1 Conclusion
48
12.2. FUTURE SCOPE CHAPTER 12. CONCLUSION AND FUTURE SCOPE
The currently proposed system not only helps to identify the disease but also provide
recommendation on that disease but it is only useful for the leaves of the citrus plant
family.
We can do some developments in this system such as system will be identied the disease
of whole citrus plant parts and also identifying the spot where the plant is infected by
the disease and also showing the percentage value that the disease is spread on plant
parts. In the future scope, we can make android which will allow farmers to click picture
of citrus leave and upload it on app and they will get infection to specic leave and exact
percentage of damage to leaves.
[5] Vinay Kukreja,Poonam Dhiman,”A Deep Neural Network based disease detection
scheme for Citrus fruits ”,IEEE Xplore Part Number: CFP20V90-ART; ISBN: 978-
1-7281-5461-9
50
Appendix A
Appendix
51
Appendix B
Base Paper
52
Appendix C
Tools Used
Overleaf : Use of Overleaf is free and open source. With a sign-up procedure, you may
create, edit, and share your projects. Overleaf is a real-time editor used for projects and
research papers. It is a LaTeX editor using the cloud to write, edit, and publish scientic
documents. Multiple users may access Overleaf at once.
Anaconda Navigator: Anaconda is an open-source tool. Anaconda Navigator also
includes a graphical user interface. It can be used for Python and R programming
languages for data science that seeks to make deployment and package management
simpler. Anaconda Navigator can launch any applications and manage the Anaconda
package without using the command line interface.
Jupiter Notebook: Jupiter Notebook is a web-based application for creating and shar-
ing any documents. Jupiter notebook is mostly used in Python programming language-
related projects. Jupiter Notebook can support programming languages such as R and
Python.
Diagrams.Net: Free online diagram software is available at Diagrams.Net..It can be
used for making owcharts, process diagrams, DFD diagrams, UML diagrams,s and net-
work diagrams.
53