10

Chapter-1

INTRODUCTION

1.1 General Introduction

Agriculture is one of the crucial tools to alleviate poverty and also helps to achieve economic

growth. Insecurity of food can create a huge risk of malnutrition. Therefore, food production is

a powerful component in the whole world. People in rural areas are mostly dependent on

agriculture. The World Bank reported that approximately 80% people in rural areas are

involved in farming.

Potatoes are a highly-successful and most favorite garden vegetable in the world. Actually, it is

a widely cultivated October-March winter crop. After rice and wheat, potato is the 3rd most

important crop in Bangladesh to enhance economic growth. Bangladesh is the 4th potato-

producing country in Asia and holds 7th position throughout the world. However, several

diseases attack potato plants during harvesting. However, several diseases affect potato

production and degrade agricultural development. Therefore, disease detection in the early

stagecan provide a better solution for a successful crop cultivation.

Therefore, detecting the conditions of potato fields and taking early treatment according to the

results could be a great solution to increase potato production, and that was the aim for doing

this study. Several traditional machine learning algorithms are already used to classify potato

leaf disease. This study combines segmentation techniques and deep learning algorithms

together to improve classification results. Image segmentation to mask the images of the potato

leaves can produce a better image dataset. Computerized pictures are transformed into different

picture sections using the method of image segmentation.

11

1.2 Problem Statement

Potato crops are vulnerable to various diseases that can cause substantial yield losses and

economic damage to farmers. Timely and accurate detection of these diseases is crucial for

implementing effective control measures and minimizing their impact. However, traditional

methods of disease detection often require specialized knowledge and laboratory testing, which

are time-consuming, expensive, and may not be readily accessible to farmers and agricultural

enthusiasts.

There is a need for a user-friendly and accessible solution that enables rapid and accurate

detection of potato diseases in the field. Existing mobile apps for plant disease detection are

limited in their scope and often lack specificity for potato diseases. Furthermore, they may rely

on cloud-based processing, which can be impractical in areas with limited internet connectivity.

The development of a native mobile app specifically designed for the detection of potato

diseases addresses these challenges. Such an app would empower farmers, agricultural experts,

and hobbyists with a convenient and efficient tool for on-the-spot disease diagnosis. By

harnessing the computational power of smartphones and integrating advanced image

recognition and machine learning techniques, the app can analyze images of potato leaves or

plants captured using the device's camera and provide real-time disease identification and

classification.

1.3 Significance of the Problem

The development of a native mobile app for the detection of potato diseases holds significant

importance and offers numerous benefits for various stakeholders in the agricultural sector. The

following points highlight the significance of addressing this problem:

1.3.1 Early Disease Detection: Timely detection of potato diseases is crucial for
12

Implementing effective control measures and preventing the spread of diseases. The
13

mobile app provides a user-friendly tool that enables farmers and agricultural

enthusiasts to identify diseases promptly, allowing them to take immediate action and

minimize crop losses.

1.3.2 Increased Accessibility: Traditional disease detection methods often require

specialized knowledge and laboratory facilities, making them inaccessible to many

farmers, particularly in remote or resource-limited areas.

1.3.3 Cost-effective Solution: Laboratory-based disease diagnosis can be expensive,

especially for farmers with limited financial resources. The native mobile app offers

a cost-effective alternative by utilizing existing smartphone technology, eliminating

the need for specialized equipment and external diagnostic services. This

affordability enables wider adoption and empowers farmers to make informed

decisions without incurring significant expenses.

1.3.4 Real-time and On-site Diagnosis: With the mobile app's real-time disease detection

capabilities, users can diagnose diseases directly in the field, eliminating the need to

transport samples to laboratories or wait for expert analysis. This immediate on-site

diagnosis allows for swift decision-making and timely implementation of control

measures, resulting in improved disease control and reduced economic losses.

1.3.5 Data-driven Agriculture: By collecting data on disease occurrences and patterns

through the mobile app, valuable insights can be gained at a larger scale. Aggregated

data can aid researchers, agronomists, and policymakers in understanding disease

dynamics, identifying regional disease hotspots, and formulating targeted strategies

for disease management and prevention.

1.3.6 Sustainable Agriculture Practices: Effective disease management contributes to

sustainable agricultural practices by reducing the reliance on chemical inputs,

14

minimizing losses, and optimizing resource utilization. The mobile app promotes
15

sustainable agriculture by providing users with the tools and knowledge to implement

integrated pest management strategies, resulting in improved crop health and reduced

environmental impact.

1.4 Objectives

The objective of this project is to develop a native mobile app specifically designed for the

detection of potato diseases. Identification and detection of plant disease through experts

observational skills are archaic ideas. This is highly costly and time-consuming. The app aims

to empower farmers, agricultural experts, and even hobbyists with an accessible and user-

friendly tool to identify and classify common potato diseases, providing them with actionable

insights and guidance for disease management.

1.4.1 Image Recognition and Classification: Image processing and analyzing on your own

is tiresome and recurrent. Computer vision and DNN played a major role in plant

disease detection. Convolutional Neural Nets are known to learn the pixel settings of

the images and the weights can be stored to test other samples with the same

accuracy. Also, it ensures high accuracy and performance in classification

1.4.2 Training Data Availability: Compiling a comprehensive and diverse dataset of potato

disease images with proper annotations and labels to train the machine learning

models. Obtaining representative samples of different disease types, severity levels,

and stages of manifestation is essential for reliable detection and classification.

1.4.3 Real-time Processing: Designing efficient algorithms that can perform image

processing and disease detection on mobile devices without relying on cloud-based

computing. This ensures the app's functionality in areas with limited or no internet

connectivity.
16

1.4.4 User Interface and Experience: Creating an intuitive and user-friendly interface that

allows users to easily capture images, initiate the disease detection process, and

receive informative outputs regarding the detected disease, including recommended

control measures and further guidance.

1.4.5 Scalability and Adaptability: Building a scalable app architecture that can

accommodate future updates, additions of new disease classes, and integration with

emerging technologies for improved disease detection and management

.
17

Convolutional Neural Network

Convolutional Neural Networks (CNNs) are widely employed for visual data classification.

They are composed of an input layer, an output layer, and multiple hidden layers. The hidden

layers typically consist of convolutional, pooling, fully connected, and normalization layers.The

basic CNN structure is shown in Figure.

Fig 2: CNN Architecture

16

Chapter-2
LITERATURE REVIEW

2.1 Summary of Papers Studied

Numerous studies have been conducted on agricultural development, which not only contribute to

economic growth but also promote a healthier environment for humans. Deep learning models and

computer vision-based approaches have gained significant attention in accelerating crop

production. In this section, we provide a concise summary of previous research. In 2017, M. Islam

et al. developed an image segmentation-based model for potato leaf detection, utilizing the Plant

Village dataset. They employed a multiclass Support Vector Machine on the segmented images,

achieving an impressive accuracy of 95% in disease classification.

Samajpati et al. introduced a hybrid model for apple disease recognition. They performed image

segmentation using k-means clustering and subsequently classified the images using the random

forest algorithm. The accuracy of their model ranged from 60% to 100%. The Plant Village dataset

has emerged as a popular resource for agricultural research, enabling researchers to analyze

agricultural data and improve production quality and quantity.

In 2021, Hassan Afzaal et al deployed the source dataset to classify early blight potato disease in a

real-time system. They utilized recent network models such as Google Net, VGG Net, and

Efficient Net. With a dataset of 5199 samples collected from four fields, the selected models

exhibited varying performance in different scenarios. The Efficient Net model achieved the

highest accuracy, ranging from 99% to 100% for two classes.

17

Kulendu et al. proposed VGG16 as an effective model for detecting early and late blight potato

leaf diseases. They initially evaluated VGG16, VGG19, MobileNet, and ResNet50 on the Plant

Village dataset, and found that VGG16 performed better after fine-tuning.

Zhou et al. introduced a hybrid deep learning model called the restructured residual dense

network, which combines the advantages of deep residual and dense networks to reduce training

time. They applied this model to the Tomato leaf dataset from AI Challenger and achieved a 95%

accuracy in identifying the 9 classes of tomato leaf diseases.

Farah Saeed et al. proposed a different approach for plant disease detection, utilizing a deep neural

model along with partial least squares (PLS) feature selection. They experimented on the Plant

Village dataset for tomato, corn, and potato crops, achieving an accuracy of 90.01%.

In another study, a MobileNet model was proposed by the author to detect potato leaf diseases.

The lightweight MobileNet V2 achieved an accuracy of 97.73% in predicting potato leaf diseases.

The Plant Village dataset was used in this study as well.

Ali Arshaghi et al. focused on the detection and classification of five different potato diseases.

They employed transfer learning techniques and achieved an accuracy of around 100%.

Md. Khalid Rayhan Asif et al. developed a CNN model to classify potato leaf images for various

diseases. They divided the dataset into two classes: normal and disorder-impacted leaves. By

applying five transfer learning algorithms, they achieved a classification accuracy of 97%.

Aditi Singh and Harjeet Kaur utilized the Plant Village dataset and applied K-means segmentation

along with SVM as the classification algorithm. They achieved an accuracy of 95.99% in detecting

potato leaf diseases.

18

Hong H et al. introduced a deep learning method for classifying tomato leaf diseases and other 8

types of disease leaves. They employed transfer learning to reduce the size of the training data,

computational time, and model complexity. Among the five deep network structures used,

Densenet_Xception achieved the highest accuracy.

Trong-Yen Lee et al. also utilized the Plant Village dataset and applied a CNN model to predict

and classify potato leaf diseases, achieving an impressive accuracy of 99.53%. Although there are

existing review papers on plant disease prediction and classification, there is still ample scope for

further research in this field, as early prediction and disease classification are crucial for enhancing

agricultural production.

El moubchiri chaimaa et al , shows the methods and processes used in a system for identifying and

classifying diseases in potato crops using image processing. However, due to the diversity of

diseases in the potato leaf, the system has also been applied to identify the crop's three most

common types of pests. Because illnesses of Downy mildew and bacterial spot are well-known

to cause harm to the plant leaf, a benefit of the detection system achieved by image processing

emphasizes the need for a design for the early detection of diseases in plants. Methods for disease

detection in potato leaves are outlined, including the five core phases involved in identifying and

labelling an object within an image, in this case, a diseased potato leaf. In the first phase, a picture

ofthe diseased leaf is acquired; in the final stage, an ANN (Artificial Neural Network) is used to

classify the image and determine whether what kind of disease it is present in the potato crop.

Finally, the processed results of the disease detection system on the potato leaf images via means

of the Mat lab program.

19

In their study, Pasalkar et al. propose a CNN-based method for predicting potato leaf diseases,

which has demonstrated promising outcomes. By harnessing the capabilities of deep learning and

computer vision techniques, they have developed an effective and efficient model for

distinguishing between healthy potato leaves and those affected by diseases like early blight and

late blight. The detection and identification of potato leaf diseases are achieved through the

utilization of image processing and machine learning algorithms. The dataset used for training and

evaluation is obtained from the Kaggle website, specifically the "new plant diseases dataset,"

which comprises diverse images of both healthy and unhealthy crop leaves. To enhance accuracy,

the researchers employ a convolutional neural network for the identification of plant diseases.

The availability of an appropriate dataset is a significant challenge in agricultural development.

While many researchers have utilized the Plant Village dataset for conducting unique experiments,

only a few have used their collected datasets, which unfortunately are not readily available for

further experimentation. In 2022, Yogeshwararao, G et al. addressed this issue by combining their

own collected data with the Plant Village dataset to predict and classify diseases in various crops.

They focused on the five most popular plants to enhance economic growth. After reviewing

previous studies, we have decided to focus our research on potato leaf disease prediction.
20

Chapter 3
REQUIREMENT ANALYSIS AND BRIEF DESCRIPTION

3.1 Description of The Project

Potato crops are susceptible to various diseases that can cause significant yield losses and

economic damage to farmers. The detection of these diseases is crucial for implementing

effective control measures. The project aims to develop a native mobile app specifically

designed for the detection of potato diseases. The app utilizes image recognition and machine

learning techniques to analyze images of potato leaves or plants captured using the

smartphone's camera. By leveraging the computational power of smartphones, the app provides

real-time disease identification and classification, empowering farmers with a convenient and

efficient tool for on- the-spot disease diagnosis. The project includes dataset compilation,

algorithm development, user interface design, and app implementation. The effectiveness of the

app is evaluated through accuracy assessments and user feedback.

Without the research, control measures can be a waste of time and money and can lead to

further plant losses. So, the farmers can use the app to identify the diseases and could be

cautious beforehand. The common diseases occurring in the plant could help the farmers in

increasing the yield. Some of the plant diseases are due to unfavorable oxygen level,

unfavorable water levels or rusts, fungus, smuts and white blisters can be responsible for

degrading the plant quality, the best way to control it is early detection for the same and then

eradicating the disease. Our real- time app could work well and detect these diseases and can

help the farmers to attenuate the crop losses and maximize the yield.
21

Dataset:
22

Potato disease detection: The dataset used for potato disease detection consists of 2152 potato

images divided into 3 categories: Early-bright, Late-bright and Healthy.

3.2 Requirement Analysis

Software Requirements:

• Operating System: Windows 10

• Platform: Python 3.11

• Server Environment: NodeJS, Fast API

• Text Editors/IDE: VS Code

• Web App: React Js

Python Libraries used:

1. Keras: It is a Python-based open-source library for neural networks. It can be used with

popular frameworks like TensorFlow, Microsoft Cognitive Toolkit, Theano, or Plaid ML. Our

team has prior experience working with it in our past projects.

2. TensorFlow: TensorFlow, specifically version 2.12, is a freely available software library

that enables dataflow and differentiable programming for various tasks. It serves as a

symbolic math library and finds extensive use in machine learning applications,

including neural networks.

3. Numpy: NumPy is a Python library that enhances the capabilities of the Python programming

language by providing support for handling large, multi-dimensional arrays and matrices.

Additionally, it offers a wide range of high-level mathematical functions that can be applied to

these arrays.
23

Functional Requirements:

1. Collection of the dataset of potato diseases.

2. Train the proposed CNN Model using this image dataset.

3. Building a web app using react Js for user interaction.

• Login/ Signup

• Real time photo capture for disease detection

• Farmer search history

4. FastAPI can be used as the backend server to handle the image processing, disease detection,

and communication with the mobile app. It can receive the image data from the mobile app

and
24

process it using image recognition and machine learning techniques. This involves loading the

image, applying preprocessing steps (if required), and utilizing trained models for disease

detection and classification.

5. Node.Js server is used for handling Login and Signup operations. It is also handling

farmer search history.

6. Updating the predicted result back to the React Js web application.

Non-Functional Requirements:

1) Accuracy: The CNN models must have good accuracy in predicting the diseases.

2) Time: The time taken by the model to predict the result should not be large enough as the

user is waiting for the result. Therefore, a perfect balance between the accuracy and the time needs

to be maintained.
25

3.2 Solution Approach

Developing a cross-platform web application for the detection of potato diseases by preparing a

bridge between the front-end side of the application and the server-side where all the CNN

models will be running, using FastAPI as server. Establishing a Node server for handling login

signup and the user history for farmers.

Advantages of using CNN:

• CNN’s use a black box technique to understand the pixels of an image. This technique is

really useful to differentiate the diseases in potatoes as most of them look alike and have

lesions.

• It is capable of giving a result with a high accuracy and in less time in case of

image classification and hence was recommended for plant identification.

• A lot of work on Object detection and Identification has been done using a CNN.

Our work is divided into five phases:

● Phase 1: Building login & signup for user registration which allows new users to

create an account.

● Phase 2: Building a website using react to input the image data from the user

● Phase 3: Building a python server that downloads the image and feeds it into the CNN

to predict the result using FastAPI.

● Phase 4: Establishing a Node.Js server for keeping track of the user search history.
25

Chapter-4

MODELING AND IMPLEMENTATION DETAILS

4.1 Design Diagram

The Design Diagram of the project is shown in Figure.

Fig 3: Design Diagram of the proposed model.

26

4.2 Implementation details and issues

4.2.1 Collection of Dataset

As a very first step, we started collecting the dataset of infected potato leaves. The most

common dataset, on which most of the current disease detection apps are based on, is the Plant-

Village dataset. The plant-village dataset consists of Potato, Tomato and Bell Pepper disease

data. Since our project is based on potato Diseases Detection. We will focus on that dataset

only.

It has 2152 images of potato leaves which can be segregated into three categories. These

categories are Early-Bright, Late-Bright and Healthy.

4.2.2 Proposed CNN Model

One significant advantage of utilizing CNNs over NNs is that you don't have to level the info

pictures to 1D as they are equipped for working with picture information in 2D. This aids in

holding the "spatial" properties of pictures. A 3-layer CNN architecture is used for generating

the pre-trained model for most of the major plant species individually and also a general single-

layered pre-trained model with all the diseases of all the plants as the classes. This has been

done to save time at the time of prediction because if the user has the knowledge about the type

of plant then running a specific pre- trained model of that plant for the disease prediction will

save time and give more accurate results.

4.2.3 Frontend

Using “React-Js” as a framework for making web application, we started developing our web
27

application screen-by-screen. Firstly, the login screen, the image upload/capture screen, then the
28

detection home screen is prepared. After adding a few more screens, things started to get

complicated and an easy way to navigate between the screens was needed. Therefore, React Router

DOM, a routing and navigation package for the react, is used to switch between the screens. A

bottom tab navigator is used at the upper layer of the app which includes the home screen, all the

detection screens and the help screen.

4.2.4 Backend

So when the user login to the website and captures/uploads the image from the image handler
screen then a request to the Fast API is generated along with the image height, image width,
and type of detection.

This Fast API consists of a CNN pre-trained saved model. This model is trained on
thepopularly known Plant Village dataset with the help of tensorflow, keras etc. As soon as
theFast API predicts the results for the input image and posts it to frontend, the frontend
displays the predicted disease and deletes that entry from the database.

At the same time, it fulfills the asynchronous call of frontend for fetching the farmer search
history.

It is very important to find the right balance between the accuracy and the time a model
takes to predict, as the user is waiting at the frontend for the result and he/she wants the result to
be accurate but doesn’t want to wait long.

Issues:

1. Our baseline CNN model follows a very simple architecture.

2. For further predicting the result with improved accuracy in much less time, a machine
with better hardware to host the server is needed.
29

4.3 Risk Analysis and Mitigation

The risk analysis and mitigation can be seen in Table 2.

Table 2: Risk Analysis and Mitigation

Risk ID Description Risk Area Probabilit Impact Risk selected Mitigation on

y
of risk for mitigation plan

1. Software System H H YES Worked on a

not and computer
responding software with better
or system failure RAM
shuts down
2. Overfitting Training M H YES Applied
of model model feature
happens using early- generation
stop Technique
function

3. Feature Software L H YES All the vectors

vector can are stored in .pkl
be corrupted file

4. Issue in Project M H YES Several time model

loading develo has been loaded
data pment

5. All paths Inefficient L H YES Setting the

not tested path counter to
recommenda minimum
tion 10.
30
29

Chapter-5

TESTING

5.1 Testing Plan

Since the project is about improving the performance of the CNN base model with newly
generated images, hence the focus of testing will be on tuning the hyperparameters and
obtaining results with different sets of images containing an increasing number of images.
The testing plan is shown in Table 3.

Table 3: Testing Plan

Type of Test Will Will Test Explanations Software

Be Component
Performed?

Requirement Yes To check the feasibility of our project in Google Colab

Testing
terms of budget, requirements, etc. this
will be done.

Unit Yes Individual units of source code will Google Colab

be run with operating procedures to
see if they are fit to use.

Integration Yes The built test cases and test data are Google Colab
integrated and then predictions are
made. The bugs (if found) are fixed,
model is re-tested.
30

Performance Yes To test whether our project will Browser, data fields
work well under expected etc.
workload, this is a must.

Stress Yes Stress Yes To compare what our Tables in the database
model predicts and what is with heavy data files.
ground truth, this test has to be
performed.

Security No Not required Not Attempted

Compliance No Not required Not Attempted

Load Yes In future, to check the Not done.

performance of our project
under real-life load, this will be
done.

Volume No Not required Not Attempted

31

Types of testing:

• Ad-hoc testing: The purpose of this testing is to find any faults, errors, and defects in our

Neural Network Models.

• Black Box testing: Since we don’t know the exact higher-order representations of images in

deeper levels of CCN, we use this type of testing to check how a given input performs with a

particular output.

• Comparison Testing: The entire results and analysis of our model are dependent upon the

comparison of base case results with Augmented results, so by comparing them both we can

make sure that the model is working fine.

• Unit Testing: Since our Model consists of multiple Neural Networks, we have to test each sub-

network individually.

5.2 Testing Results

Below Figure shows the plot of loss function and accuracy.

Snapshot of training the model 4: Plot of training accuracy and Validation Accuracy
and loss.
32

5.3 Error and Exception Handling

The errors that we encountered were in following domains:

• Image conversion to numpy files

• Image resolution(shape)

Although we solved all these errors, we did encounter some exceptions such as a few

multichannel images

5.4 Limitations

There are certain limitations and challenges of the solution as below:

• A vast variety of flora and fauna are present therefore, it becomes difficult to identify all
the flowers accurately.

• The variations in diseases in the plant also make it difficult to effectively predict it.
• Some large similarities that occur in nature make it tough to maintain efficacy in the model.
• There are very few datasets available in this field. This hinders our model to be
rigorously tested upon various categories of people.
33

Chapter-6
FINDINGS, CONCLUSION AND FUTURE WORK

6.1 Findings

The accuracy of diseases for each of the species is shown in table 5.

Table 4: Accuracy of detection of disease in each plant

Accuracy 97.0 %

Loss 93.4 %

Here the general model including all the diseases of all the plants is trained only on the single-
layered CNN model for quicker results. The accuracy of the classification of the flowers is 90.3
% when trained over the 3 layered CNN model.

The training of models for the potato diseases is shown in Fig 5. Fig 6 and Fig 7 shows the
login and signup. Fig. 8 and 9 shows the UI for uploading and submitting the plant image for
detection. Fig. 10 shows the predicted results for the uploaded plant leaf. Fig.11 shows the UI
for the farmer search history.
34

Fig 5: Snapshot of training the model

Fig 6: Snapshot of the Login Page

35

Fig 7: Snapshot of the Signup Page

Fig 8: UI for uploading image

36

Fig 9: UI for submitting image

Fig 10: Snapshot of Predicted Results

37

Fig 11: UI for Farmer Search History

38

6.2 Conclusion
In this project, an automated system is developed with React Js that predicts the diseases in

potato plants. The algorithm is validated for 3 categories namely early blight, late blight and

healthy. The general accuracy for disease detection is 97 % while the reported results are very

encouraging. The detection works on the “Plant Village dataset ” of a total of 2152 images of

potato leaves. The CNN pre-trained model is saved and used directly for predictions in the

backend to save time . The results obtained until now are encouraging with good accuracy but

still, there is a scope of increasing the efficiency by using various other machine learning

models.

6.3 Future Work

Traditionally, methods of potato disease identification were developed using different models,

but these models could not cover all the disease. In our model, we have tried to cover major

plants and their diseases but the number of flora and fauna species is vast and developing a

model for all the species is very difficult but we’ll try to add new models for different plant

species or flower species in the near future. Secondly, the backend python server which runs

CNN models to predict the result needs to be hosted online. Currently, the backend python

server is running on a local device. We would also convert the existing website into a dedicated

mobile app for potato disease detection for better and more accessible user interaction.

We would continuously update and expand the disease database used in our web app and

increase our coverage of disease detection to other plants. We would also try to incorporate

emerging and lesser-known diseases to ensure comprehensive disease identification and

classification. We will also integrate geolocation data and climate information into our website

to enhance disease prediction.

39

REFERENCES

1) Ahmad, Saeed & Jasim, Ihsan & Nosher, Azmat Ali & Rehman, Zubaida & Shah, Ali & Chaimaa,
El & Sabugaa, Michael & Jassim, Moayad & Mehr, Pir. (2023). “Modern image processing
techniques for identifying and classifying Disease Detection in Potatoes Crop”.
INTERNATIONAL JOURNAL OF ENGINEERING RESEARCH & TECHNOLOGY (IJERT)

2) Deep Kothari , Harsh Mishra , Vishal Pandey , Mihir Gharat, Rashmi Thakur, 2022, “Potato
Leaf Disease Detection using Deep Learning”, INTERNATIONAL JOURNAL OF
ENGINEERING RESEARCH & TECHNOLOGY (IJERT)

3) JAYASHREE PASALKAR, GANESH GORDE*, CHAITANYA MORE, SANKET MEMANE

And VAISHNAVI GAIKWAD ,(2023), “Potato Leaf Disease Detection using Machine
Learning”. INTERNATIONAL JOURNAL OF ENGINEERING RESEARCH &
TECHNOLOGY (IJERT)

4) Garcia R, Martinez L, Perez J, Rodriguez M, Sanchez A. "Web-based platform for early detection
of potato diseases using image processing techniques." INTERNATIONAL JOURNAL OF
ENGINEERING RESEARCH & TECHNOLOGY (IJERT)

5) Patel R, Gupta S, Sharma V.(2015) "Web-based tool for potato disease identification using
artificial intelligence." International Journal of Engineering Research & Technology.

6) Nguyen T, Trinh H, Le Q, Pham T, Doan H. (2014) "A web application for potato disease
detection using ensemble learning algorithms." International Journal of Engineering Research &
Technology.

7) Shruthi U., Nagaveni V. and Raghavendra B. K. 2019 A Review on Machine Learning

Classification Techniques for Plant Disease Detection. International Journal of
Engineering Research & Technology.

8) Aditi Singh and Harjeet Kaur 2021 IOP Conf. Ser.: Mater. Sci. Eng. 1022 012121 , “Potato plant
disease detection And classification using ML Methodologies”. International Journal of
Engineering Research & Technology.

9) M. I. Tarik, S. Akter, A. A. Mamun and A. Sattar,(2021), "Potato Disease Detection Using Machine
Learning," 2021 Third International Conference on Intelligent Communication Technologies and
Virtual Mobile Networks (ICICV).
40

10) Dor Oppenheim, Guy Shani, Orly Erlich, and Leah Tsror (2019), “Using Deep Learning for Image-
Based Potato Tuber Disease Detection”. INTERNATIONAL JOURNAL OF ENGINEERING
RESEARCH & TECHNOLOGY (IJERT).

11) Asif, Md. Khalid & Rahman, Asfaqur & Hena, Most. Hasna. (2020). CNN based Disease
Detection Approach on Potato Leaves. INTERNATIONAL JOURNAL OF ENGINEERING
RESEARCH & TECHNOLOGY (IJERT)

12) Lee, Hagyeong & Song, Jongwoo. (2019). “Introduction to convolutional neural network using
Keras; an understanding from a statistician. Communications for Statistical Applications and
Methods.” INTERNATIONAL JOURNAL OF ENGINEERING RESEARCH &
TECHNOLOGY (IJERT)

13) Afzaal, Aitazaz A. Farooque ,Arnold W. Schumann ,Nazar Hussain ,Andrew McKenzie-Gopsill
3ORCID,Travis Esau 4,Farhat Abbas and Bishnu Acharya (2021). “Detection of a Potato Disease
(Early Blight) Using Artificial Intelligence”

14) C. Zhou, S. Zhou, J. Xing and J. Song, (2021). "Tomato Leaf Disease Identification by
Restructured Deep Residual Dense Network," INTERNATIONAL JOURNAL OF
ENGINEERING RESEARCH & TECHNOLOGY (IJERT)

15) Chen, W.; Chen, J.; Zeb, A.; Yang, S.; Zhang, D. (2022) “Mobile convolution neural network for
the recognition of potato leaf disease images. INTERNATIONAL JOURNAL OF
ENGINEERING RESEARCH & TECHNOLOGY
41

APPENDIX-I
PLAGIARISM REPORT
42