Plant Disease Detection by Using Image

Processing
A project report stage I submitted in partial fulfillment of the
requirements for the degree of Bachelor of Computer Science

Submitted by

Nagesh Avhad

Under the guidance of

Prof. Mrs.Poonam Dubey

Department of Computer Science G. R. Patil College of Science,

Commerce and Arts, Dombivali

University of Mumbai
2023-2024
 G. R. Patil College of Science, Commerce and Arts,
Dombivali
CERTIFICATE
This is to certify that the project entitled

Plant Disease Detection by Using Image

Processing

Submitted by

Nagesh Avhad

is successfully completed for the Bachelor of Computer Science

as prescribed by University of Mumbai.

Internal External
 ABSTRACT
Agriculture is a field that significantly impacts the lives and economic status
of human beings. Improper management can result in a decline in the qual-
ity of agricultural products. Farmers often lack knowledge about diseases,
thereby affecting their crops and leading to reduced production. Although
Kisan call centers are available, they do not operate 24/7, and communica-
tion failures can occur.
When farmers are unable to adequately explain a disease over the phone,
there is a need to analyze images of the affected areas. While images and
videos of crops offer a better perspective, many farmers are unaware that
agro-scientists can provide effective solutions to ensure the health of crops.
It’s crucial to note that if crop productivity is compromised, it poses a signifi-
cant risk to providing good and healthy nutrition. Fortunately, advancements
in technology have led to the development of smart devices capable of rec-
ognizing and detecting plant diseases. Prompt recognition of illnesses can
facilitate faster treatment, minimizing the negative impacts on the harvest.
Contents

1 Introduction 1
1.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Literature Survey 3

3 Implementation 6
3.1 Existing System . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.1 Disadvantages on existing system . . . . . . . . . . . . 6
3.1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . 7
3.1.3 Focus on Plant Disease Detection with Image Processing 8
3.1.4 Proposed System . . . . . . . . . . . . . . . . . . . . . 8

4 Methodology 11
4.1 Training Phases . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 CNN Architecture . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.3 Analysis and Design Consideration . . . . . . . . . . . . . . . 15
4.3.1 Use Case Diagram . . . . . . . . . . . . . . . . . . . . 15
4.3.2 Sequence Diagram . . . . . . . . . . . . . . . . . . . . 16
4.3.3 Activity Diagram . . . . . . . . . . . . . . . . . . . . . 17

5 Development Tools 18
5.1 Hardware and Software requirements . . . . . . . . . . . . . . 18

6 Further Work 19

7 Conclusion and future work 20

References 21

ii
List of Figures

3.1 Proposed System . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.1 Training Phases . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.2 CNN Architecture . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.3 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . . 15
4.4 Sequence Diagram . . . . . . . . . . . . . . . . . . . . . . . . 16
4.5 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . 17

iii
Chapter 1

Introduction

Agriculture holds a crucial role in our country, with approximately 70% of the
population relying on farming for their livelihood. The distressing reality of
farmers resorting to suicide due to production losses is a grave concern. This
issue can be mitigated to some extent by incorporating new technologies that
empower farmers to enhance their harvesting practices. Plants have emerged
as vital energy sources and play a pivotal role in addressing environmental
challenges like global warming. Numerous diseases pose threats to plants,
capable of causing severe economic, social, and ecological consequences. De-
tecting plant diseases proves challenging as some remain invisible to the naked
eye or manifest only when it’s too late to take action. Traditional methods
for identification and detection are laborious and time-consuming.
A prevailing strategy involves leveraging remote sensing techniques that uti-
lize multi and hyperspectral image captures. Approaches employing digital
image processing tools are commonly adopted to attain their objectives. No-
tably, image processing technology in agricultural research has undergone
significant development. The advent of deep learning techniques, especially
Convolutional Neural Networks (CNNs), has propelled substantial advance-
ments in image processing. Since 2016, numerous applications for the au-
tomatic identification of crop diseases have been developed. These applica-
tions serve as a foundation for the creation of expert assistance or automated
screening tools, fostering more sustainable agricultural practices and bolster-
ing food production security.

1.1 Objective
To address the detection of plant leaf diseases and pests among inexperienced
farmers, we propose developing an Mobile application. This application is

1
specifically designed to cater to the needs of young farmers, offering a user-
friendly interface for disease detection in plants. Through the Mobile app,
farmers can easily access the application on their mobile devices while work-
ing in their fields. The app will not only identify plant diseases but also
provide suitable remedies based on the detected issues. This shift to an
Mobile application ensures accessibility and convenience, empowering a new
generation of farmers with a practical tool for effective crop management.

1.2 Problem Statement

Develop an mobile application that utilizes a Constitutional Neural Network
(CNN) to construct a model based on images of plant leaves. The primary
goal is to detect diseases and recommend suitable remedies to counteract
them. In instances where a leaf is identified as unhealthy or diseased, the
application will specify the type of disease afflicting the leaf. The overarching
objective is to classify whether the leaf exhibits conditions such as Bacterial
spot, Late Blight, or Yellow Curl Leaf virus, employing a machine learning
algorithm like Logistic Regression.

2
Chapter 2

Literature Survey

1. Smart Farming: Pomegranate Disease Detection Using Image

Processing, 2015.
A web-based tool was created to detect fruit diseases through
the submission of fruit images to the system. Feature ex-
traction was performed using parameters such as color,
morphology, and CCV (color coherence vector). The k-
means algorithm was employed for clustering, and SVM was
utilized for classifying the fruit as infected or non-infected.
This study achieved an accuracy rate of 82% in identifying
pomegranate diseases.(from reference [1])
2. Leaf Disease Detection and Recommendation of Pesticides us-
ing Convolution Neural Network, 2018.
Crop production issues are widespread in India, signifi-
cantly affecting rural farmers, the agriculture sector, and
the overall national economy. The condition of leaves in
crops is pivotal as it provides crucial information about
the anticipated quantity and quality of agricultural yield.
This paper proposes a system that operates on the pre-
processing and feature extraction of leaf images from a
plant village dataset. It employs a Convolution Neural
Network (CNN) for the classification of diseases and rec-
ommends pesticides using TensorFlow technology.
The key components of our system involve an Android ap-
plication with Java Web Services (JWS) and Deep Learn-

3
 ing. We utilized a Convolution Neural Network with vary-
ing layers—five, four, and three—to train our model. The
Android application serves as the user interface with JWS
facilitating interaction between these systems. Our results
indicate the highest accuracy achieved by the 5-layer model,
reaching 95.05% for 15 epochs, and the highest validation
accuracy for the 5-layer model, attaining 89.67% for 20
epochs using TensorFlow.(from reference [2])
3. Plant Disease Detection using CNN.
Plant diseases pose a persistent challenge for smallholder
farmers, posing threats to both income and food security.
The increasing prevalence of smartphones and advance-
ments in computer vision models present an opportunity
for employing image classification techniques in agricul-
ture. Convolutional Neural Networks (CNNs), recognized
as state-of-the-art in image recognition, offer the potential
for swift and accurate disease diagnosis. This literature re-
view explores recent developments in leveraging CNNs for
plant disease detection.(from reference [3])
4. Improvement of plant disease classification accuracy with gen-
erative model-synthesized training datasets.
Digitalization in agriculture requires critical research into
applications of artificial intelligence across various special-
ized domains. This study aims to investigate the applica-
tion of image synthesis technology to address the data vol-
ume constraints in achieving accurate digital plant disease
phenotyping. We conducted an experiment using a deep
convolutional generative adversarial network (DC-GAN) to
synthesize photorealistic data for both healthy and bac-
terial spot disease-infected tomato leaves. The training
dataset comprised 1,272 instances per class. Furthermore,
we employed a 3-block Visual Geometry Group (VGG) con-
volutional neural network (CNN) model with dropout reg-
ularization and a single epoch to compare classification ac-

4
curacies between the original dataset and various synthetic
datasets. Our results demonstrated that the third DC-
GAN synthesized training dataset, consisting of 3,816 syn-
thetic examples for both healthy and bacterial spot-infected
tomato leaf classes, outperformed the original training dataset,
which contained 1,272 real examples for both tomato leaf
classes. Specifically, the accuracy with the former dataset
on a 3-block VGG CNN model with dropout regularization
and one epoch was 77.088%, compared to 76.447% accu-
racy with the latter dataset on the same classifier.(from
reference [4])

5
Chapter 3

Implementation

3.1 Existing System

The current approach to plant disease detection predominantly relies on man-
ual observation and expert analysis. In this traditional method, farmers
play a crucial role in identifying symptoms such as discoloration, lesions, or
unusual patterns in plant growth. Once symptoms are observed, farmers
typically consult agricultural experts for guidance and diagnosis. However,
this process is inherently subjective, as it heavily depends on the observer’s
level of expertise. The identification of diseases through manual observation
is time-consuming and can lead to delayed responses, impacting crop yields
and overall agricultural productivity.
Moreover, the reliance on human expertise poses challenges, especially in
cases where diseases manifest in subtle ways or resemble nutritional deficien-
cies. As a consequence, misdiagnoses are not uncommon, and the lack of
timely and accurate information may result in ineffective treatment strate-
gies. Recognizing the limitations of this system, there has been a growing
interest in integrating technological solutions to enhance the efficiency and
precision of plant disease detection.

3.1.1 Disadvantages on existing system

1. Prone to Human Error:

ˆ The manual approach is susceptible to inaccuracies due to human

error.
ˆ Farmers may not consistently and accurately identify or interpret
disease symptoms.

6
 2. Delayed Responses:

ˆ The traditional method often results in delayed responses to dis-

ease outbreaks.
ˆ Delays can allow diseases to progress, leading to potential irre-
versible damage to crops.

3. Reliance on Expert Knowledge:

ˆ The system heavily depends on the availability of agricultural ex-

perts.
ˆ Regions with limited access to agricultural expertise face a bot-
tleneck in disease diagnosis.

4. Subjective Nature of Manual Observation:

ˆ Manual observation is inherently subjective.

ˆ Establishing consistent and standardized diagnostic criteria is chal-
lenging.

5. Lack of Standardization:

ˆ The subjective nature hinders the establishment of consistent di-

agnostic criteria.
ˆ Lack of standardization impedes effective communication and col-
laboration among stakeholders.

6. Resource Intensive:

ˆ Expert intervention is necessary for accurate disease diagnoses.

ˆ This reliance on experts implies higher costs and resource require-
ments for farmers.

3.1.2 Problem Statement

The existing system for plant disease detection, primarily reliant on manual
observation and expert analysis, presents significant challenges in accuracy,
timeliness, and resource efficiency. The manual approach is prone to human
error, leading to inconsistencies in identifying disease symptoms by farm-
ers. This subjectivity, coupled with delayed responses inherent in traditional
methods, allows diseases to progress, causing potential irreversible damage

7
to crops. Furthermore, the dependence on agricultural expertise creates bot-
tlenecks, particularly in regions with limited access to such knowledge.
The subjective nature of manual observation complicates the establishment of
standardized diagnostic criteria, hindering effective communication and col-
laboration among farmers, researchers, and agricultural professionals. More-
over, the need for expert intervention implies higher costs and resource re-
quirements for farmers seeking accurate disease diagnoses. In light of these
disadvantages, there is a pressing need for a more advanced and technologically-
driven solution for plant disease detection

3.1.3 Focus on Plant Disease Detection with Image

Processing
In particular, the integration of image processing and machine learning tech-
nologies offers a promising avenue for improvement. By leveraging these
advancements, the agricultural community aims to address the limitations
of the current system. The goal is to enhance accuracy, provide timely re-
sponses to disease outbreaks, and reduce the dependency on human expertise.
Standardized diagnostic tools powered by image processing algorithms can
streamline the identification process, promoting consistency and facilitating
collaboration among stakeholders. Additionally, such technological solutions
have the potential to be more accessible, even in remote areas, democra-
tizing disease detection capabilities and contributing to improved resource
efficiency for farmers.

3.1.4 Proposed System

In response to the limitations of the existing plant disease detection system,
a novel approach leveraging image processing and machine learning is pro-
posed. This advanced system aims to enhance the accuracy, timeliness, and
accessibility of disease diagnosis. Through the integration of a mobile app,
farmers can conveniently upload images of plant leaves, enabling automated
analysis of disease symptoms. The system’s image processing algorithms will
provide a more objective and standardized method for identifying diseases,
reducing the impact of human error. The real-time nature of the proposed
system ensures timely responses to disease outbreaks, mitigating potential
crop damage. Moreover, by decentralizing the diagnostic process and re-
ducing the reliance on expert knowledge, the mobile app empowers farmers,
especially in regions with limited access to agricultural expertise. This in-
novative solution represents a significant step towards a more efficient and

8
inclusive plant disease detection system, fostering better communication and
collaboration among agricultural stakeholders.

Figure 3.1: Proposed System

Advantages on Proposed system

1. Enhanced Accuracy:
ˆ The integration of image processing algorithms ensures a more
accurate and objective identification of plant diseases, reducing
the impact of human error in manual observation.
2. Timely Responses:
ˆ The real-time capabilities of the proposed system allow for swift
responses to disease outbreaks. Early detection through mobile
app image uploads helps farmers implement timely interventions,
preventing further crop damage.
3. Accessibility and Inclusivity:
ˆ The mobile app provides a user-friendly interface, making disease
detection accessible to a wider range of farmers, including those
in remote areas. This inclusivity democratizes the benefits of ad-
vanced technology in agriculture.
4. Standardized Diagnoses:
ˆ The use of image processing technologies contributes to standard-
ized diagnostic criteria, promoting consistency in identifying and
categorizing plant diseases. This facilitates effective communica-
tion and collaboration among farmers and agricultural experts.

9
5. Reduced Dependency on Expertise:

ˆ By enabling farmers to upload images for automated analysis, the

proposed system reduces the dependency on expert knowledge.
This is particularly advantageous in regions with limited access
to agricultural expertise, allowing for more independent decision-
making by farmers.

6. Cost-Efficiency:

ˆ The proposed system minimizes the need for on-site expert consul-
tations, reducing associated costs for farmers. Automated image
analysis via the mobile app offers a cost-effective alternative to
traditional, resource-intensive diagnostic processes.

7. Empowerment of Farmers:

ˆ The mobile app empowers farmers by giving them a proactive role

in disease detection and management. It enhances their capacity
to make informed decisions about crop health, fostering a sense of
ownership and control over their agricultural practices.

8. Facilitation of Research and Data Collection:

ˆ The aggregated data from mobile app uploads can contribute to

broader research efforts in understanding regional disease pat-
terns. This wealth of data aids in the development of more ef-
fective and region-specific disease management strategies.

9. User-Friendly Interface:

ˆ The mobile app’s intuitive design ensures that farmers can easily
upload images and receive prompt feedback. The user-friendly
interface contributes to the overall usability and acceptance of the
system.

10. Environmental Sustainability:

ˆ By facilitating early disease detection and targeted interventions,

the proposed system supports sustainable agricultural practices.
This can lead to optimized pesticide use, reducing environmental
impact and promoting eco-friendly farming methods.

10
Chapter 4

Methodology

4.1 Training Phases

The main aim is to design a system that is efficient and provides disease
names and pesticide names as fast as possible. For that purpose, we use
two phases: the 1st is the training phase, and the 2nd is the testing phase.
In the 1st phase: Image acquisition, Image Pre-processing, and CNN-based
training. In the 2nd phase, Image acquisition, Image Pre-processing, Classi-
fication, disease identification, and suggesting appropriate remedies.

1. Image Acquisition: For training, an image is taken from a

file on the system. For the purpose of testing, the images
of the plant leaves are captured as and when required and
then transferred to a folder on the system for analysis.
2. Image Pre-Processing: Image should be processed before
sending to the algorithm for testing and training purpose.
For that purpose, in this project, the image is scaled or
resized into a 50 x 50 dimension. Since we used a color
image, we don’t need any color conversion techniques, and
that preprocessed image is directly passed to the algorithm
for training and testing purposes
3. Convolutional Neural Network: Once pre-processing is com-
plete, the Convolutional Neural Network (CNN) is em-
ployed for training purposes, resulting in a trained model.
This CNN method is implemented with the assistance of
TensorFlow. Using this model, we classify the image ob-
tained after the pre-processing of the testing image. Sub-

11
 Figure 4.1: Training Phases

sequently, we obtain the specific disease name or determine

the healthy status of the leaf if no disease is present. With
the identified disease name, we retrieve remedies that can
aid the farmer in taking action to eradicate or mitigate the
effects of the disease.
4. Training: Training a network is a procedure of obtaining
kernels in convolution layers and weights in fully connected
layers that reduce differences on a training dataset between
output predictions and specified ground truth labels. In our
work, we used 87% of the data for training. Through this
stage, the network that has been built learns by extracting
features from plant leaf disease images to distinguish each
image based on these learned features.
5. Detection for Plant Leaf Diseases and Remedies Prediction:
After the previous operations, plant species diseases are
detected and classified into three types of plant diseases,
namely bacterial spots, late blight, and yellow curl leaf
virus. After detection and classification are completed, the
prediction of remedies is carried out by the application.

12
4.2 CNN Architecture
1. Input Layer: The input layer contains the input images and
their pixel values.
2. Convolution Layer: The main building block of a CNN is the
Convolutional layer. It involves the straightforward appli-
cation of filters of various sizes to the input image, resulting
in activation. The repeated use of the input image with the
same filter produces a map in the activation called a fea-
ture map. This map indicates the strength of the detected
features in the input image. Each feature map is convo-
luted with the help of numerous input feature graphs in
Convolutional Neural Networks.
3. Pooling Layer: A pooling layer often follows a convolutional
layer and can be utilized to depreciate the dimensions of
feature maps and parameters of the network. Pooling lay-
ers are also invariant in interpretation, similar to convolu-
tional layers, because their calculations take into account
neighboring pixels. The most widely used approaches are
average pooling and max pooling. In our research, we used
a max-pooling layer.
4. Non-Linear Layer: A non-linear transformation is applied
to the input by the CNN, the objective of which is to clas-
sify the features within each hidden layer. In the CNN
structure, we use Rectified Linear Units (ReLU). Rectified
Linear Units are commonly used as a nonlinear transforma-
tion. This type of layer executes a simple operation with
a threshold, where any input value smaller than zero is set
to zero.
5. Fully Connected Layer: The data arrives at the last layer
of the CNN, which is the fully connected node, after much
iteration of the prior layers. In the two neighboring layers,
the neurons are connected directly to the neurons within
the fully connected network, as shown in Fig.
6. Normalize Layer: In our proposed system, we use a batch
normalization layer. The batch normalization layer nor-
malizes each channel through a mini-batch. This can help
decrease sensitivity to data variations.

13
7. Softmax Layer: The network’s performance can be difficult
to interpret. It is customary to conclude the CNN with a
Softmax function in classification issues. After extracting
values for 15 classes of plant diseases in the fully connected
step, a Softmax will be applied to them, so that the class
will be selected in each process based on the features that
were extracted through the previous layers that the images
of plant diseases went through. In this layer, the correct
class of disease is determined by applying the Softmax func-
tion.

Figure 4.2: CNN Architecture

14
4.3 Analysis and Design Consideration
4.3.1 Use Case Diagram

Figure 4.3: Use Case Diagram

15
4.3.2 Sequence Diagram

Figure 4.4: Sequence Diagram

16
4.3.3 Activity Diagram

Figure 4.5: Activity Diagram

17
Chapter 5

Development Tools

5.1 Hardware and Software requirements

ˆ Software Requirement:

– Windows/Linux/Mac OS
– Java
– Python
– NodeJS
– Ionic
– IDE
* Jupyter Notebook
* VS Code

ˆ Hardware Requirement:

– Smart Phone
– Laptop/Desktop with
* i5 processor
* 4GB RAM

18
Chapter 6

Further Work

ˆ Design and develop Ionic mobile App.

ˆ Collect sample dataset for training.

ˆ Implement training logic and train model for disease detection.

ˆ Test application with White box and black box Testing.

19
Chapter 7

Conclusion and future work

Plant Disease and Detections using Image Processing is still in the develop-
ment phase. Some enhanced functions are still needed to be added continu-
ously. In future work in this area, it plans to explore the following extensions.

ˆ Improvement in user interfaces in android device as well as in webserver.

ˆ Improving the Training accuracy by enhancing CNN training.

Hence drawback with the existing system which is mostly manual work
will be overcome by the proposed system.

20
Bibliography

[1] Manisha Bhange and H.A. Hingoliwala. Smart Farming: Pomegranate

Disease Detection Using Image Processing, 2015

[2] Pranali K. Kosamkar, V.Y. Kulkarni, Krushna Mantri, Shubham

Rudrawar, Shubhan Salmpuria and Nishant Gadekar. Leaf Disease De-
tection and Recommendation of Pesticides using Convolution Neural
Network, 2018

[3] M. Shobana, Vaishnavi S, Gokul Prasad C, Pranava Kailash SP, Mad-

humitha K P, Nitheesh C and Kumaresan N. Plant Disease Detection
using CNN, 2022

[4] Enow Takang Achuo Albert, Ngalle Hermine Bille and Eddy L Ngonkeu.
Improvement of plant disease classification accuracy with generative
model-synthesized training datasets, 2023

21
 Acknowledgement
This work was influenced by countless individuals whom we were fortunate
enough to meet during our phase-I of the project duration, while space does
not permit us to acknowledge them all, we would be remiss if we did not
acknowledge the following individuals whose guidance, support and wisdom
so greatly influenced this work.

n particular, we would like to thank our Project Guide Prof. Mrs.Poonam

Dubey for her moral support and guidance to complete our Phase-I of
Project. Further we also wish to extend our thanks to all the staff mem-
bers as well as other colleagues for attending our Seminars and for their
insightful comments and constructive suggestions to improve the quality of
this Project work.

Date: