A

PRELIMINARY PROJECT REPORT

ON

Custard Apple Disease Prediction with Deep Learning

SUBMITTED TO THE SAVITRIBAI PHULE PUNE UNIVERSITY, PUNE

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE AWARD OF THE DEGREE OF

BACHELOR OF ENGINEERING
INFORMATION TECHNOLOGY

BY

Ashutosh Patil B1900508513

Gayatri Bhangle B1900508532
Sejal Kadam B1900508597
Poonam Vetal B1900508731

Under the guidance of

Mrs. P. S. Shinde

Department Of Information Technology

Pune Institute of Computer Technology
Pune - 411 043.
2024-2025
 SCTR’s PUNE INSTITUTE OF COMPUTER TECHNOLOGY
DEPARTMENT OF INFORMATION TECHNOLOGY

CERTIFICATE

This is to certify that the preliminary project report entitled

A Systematic Review on Fruit Disease Detection Using Deep Learning
submitted by
Ashutosh Patil B1900508513
Gayatri Bhangle B1900508532
Sejal Kadam B1900508597
Poonam Vetal B1900508731
is a bonafide work carried out by them under the supervision of Mrs. P.
S. Shinde and it is approved for the partial fulfillment of the requirement of
Savitribai Phule Pune University for the award of the Degree of Bachelor of
Engineering (Information Technology).

This project report has not been earlier submitted to any other Institute or
University for the award of any degree or diploma.

Mrs. P. S. Shinde Dr. A. S. Ghotkar

Project Guide HOD IT

Dr. S. T. Gandhe
SPPU External Guide Principal

Date:
Place:

i
 Acknowledgement

The purpose of the acknowledgement page is to express my gratitude to those who helped
with the thesis/other tasks and responsibilities related to the writing of the report. Ev-
eryone who helped with the process received training in the process and expressed an
interest in the process. Therefore, it is necessary to express interest in shorthand rather
than explaining effort and thought. This suggestion is not recommended for the rest of
the report. Remember to use formal language, even when thanking family and friends.
, professionals, supervisors, organizational leaders, companies, family and friends. This
is usually the name of a person and the type of assistance you received. For example,
a consultant who helped you design a workshop , a person who helped you complete a
building or workshop program, a person who helped you learn how to use a computer, a
person who provided raw materials for the workshop ¿ workshop, etc.

Ashutosh Patil B1900508513

Gayatri Bhangle B1900508532
Sejal Kadam B1900508597
Poonam Vetal B1900508731

ii
 Abstract

Detection of plant diseases is an important task in today’s agriculture, and early detection
and treatment can reduce crop losses and increase yields. Among the studied fruit vari-
eties, custard apple (Annona squamosa) has received little attention in disease research
despite its economic and agricultural importance. This review presents a comprehensive
review of existing fruit disease detection methods, focusing on deep learning and its appli-
cation in cherry apple disease detection. Deep learning models, including Convolutional
Neural Networks (CNN), VGG, DenseNet, etc., have been quite successful in identifying
and classifying diseases based on leaf patterns. This paper examines the performance,
advantages, and limitations of these models on different datasets. It also highlights issues
such as lack of information, class bias, and environmental change, and presents strategies
such as knowledge support and learning to overcome these issues. There are significant
differences in research in the literature, highlighting the need for more research in this
area. Ultimately, the aim is to improve disease diagnosis for cherimoya crops and con-
tribute to sustainable agriculture.

iii
Contents

Certificate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Sponsorship letter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Literature Survey 3

3 Requirement Specification and Analysis 4

3.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.4 Proposed Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.5 Project Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.5.1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.5.2 Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . 7
3.5.3 Non Functional Requirements . . . . . . . . . . . . . . . . . . . . 7
3.5.4 Hardware Requirements . . . . . . . . . . . . . . . . . . . . . . . 8
3.5.5 Software Requirements . . . . . . . . . . . . . . . . . . . . . . . . 8
3.6 Project Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.6.1 Project Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.6.2 Module Split-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.6.3 Project Team Role and Responsibilities . . . . . . . . . . . . . . . 10
3.6.4 Project Plan 3.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

iv
 3.6.5 PERT Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4 System Analysis and Design 11

4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Necessary UML Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2.1 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2.2 DFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2.3 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2.4 Class Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

5 Implementation 17
5.1 Stages of Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.1.1 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.1.2 Implementation of Modules . . . . . . . . . . . . . . . . . . . . . 17
5.2 Experimentation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

6 Conclusion and Future Scope 20

6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 Limitations of the Project . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.3 Future Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

References 22

Plagiarism Report 24

Base Paper 24

Review Sheets 25

Monthly Planning Sheet 26

Project Achievements 27

v
List of Figures

4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.2 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.3 Data Flow Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.4 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.5 Class Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

vi
List of Tables

2.1 Literature Survey on Fruit Disease Detection . . . . . . . . . . . . . . . . 3

3.1 PERT Table for Project Tasks . . . . . . . . . . . . . . . . . . . . . . . . 10

vii
 Abbreviations
CNN : Convolutional Neural Network

VGG : Visual Geometric Group

IoT : Internet of Things

DL : Deep Learning

viii
 Custard Apple Disease Prediction with Deep Learning

1. Introduction

1.1 Introduction
In recent years, the application of deep learning has revolutionized various fields, includ-
ing agriculture. Fruit disease detection is a critical area in precision agriculture, where
early identification of diseases can prevent crop loss, improve productivity, and promote
sustainable farming practices. Traditional methods for diagnosing plant diseases rely
heavily on human expertise, which can be subjective, time-consuming, and prone to er-
ror. To overcome these limitations, automated detection methods using deep learning
have gained prominence, providing more accurate and efficient solutions.
Deep learning algorithms, particularly Convolutional Neural Networks (CNNs), have
shown remarkable success in image-based disease detection by analyzing symptoms such
as color changes, texture, and shape distortions in fruit leaves and other plant parts.
Various studies have applied these techniques to a wide range of fruits, including ap-
ples, bananas, and grapes, achieving significant improvements in disease identification
accuracy.
This systematic review aims to provide a comprehensive overview of in-depth stud-
ies on fruit disease detection, focusing on the main methods, datasets and their results.
The review also highlights inconsistencies in current research, particularly on Annona
squamosa, a fruit whose disease diagnosis has not yet been investigated. By present-
ing information from the literature, this review provides a basis for future research and
progress in this development.

1.2 Motivation
The increasing global demand for quality fruit production, coupled with the impact of
diseases on crop yield and economic losses, necessitates efficient and reliable disease de-
tection methods. Traditional manual inspection of fruit diseases is often labor-intensive,
subjective, and prone to inaccuracies, especially in large-scale farming.
Deep learning offers a promising solution by automating the detection process with
high accuracy, helping farmers identify diseases early and take timely action. Despite
significant progress in this area, certain fruits like custard apple (Annona squamosa) have
received little attention in the research community, leaving a gap in disease detection

PICT,Pune 1 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

methods for this agriculturally valuable fruit. This paper seeks to address this gap,
providing an in-depth review of deep learning applications in fruit disease detection and
laying the groundwork for future advancements in the field.

1.3 Objectives
The primary objective of this systematic review is to analyze and synthesize existing
research on the application of deep learning techniques in fruit disease detection, with
a particular focus on identifying gaps in the literature, especially for under-researched
fruits like custard apple (Annona squamosa). This study aims to:

• Examine the effectiveness of various deep learning models (e.g., CNN, VGG, DenseNet)
in detecting and classifying fruit diseases.

• Identify the challenges faced in implementing these models, including data require-
ments, environmental variability, and practical field deployment.

• Highlight the underrepresentation of custard apple in current research and pro-

pose future directions for developing deep learning-based disease detection systems
tailored to this fruit.

• Explore how deep learning technologies can enhance early disease detection, improve
yield quality, and support sustainable agricultural practices.

1.4 Scope
This systematic review focuses on the use of deep learning techniques for fruit disease
detection, covering a range of fruits but emphasizing the under-researched custard apple.
The review will explore various deep learning models such as CNN, VGG, and DenseNet,
assessing their accuracy and applicability. It also addresses challenges like data availabil-
ity, environmental factors, and practical field usage. The study aims to provide insights
into improving disease detection methods and identify areas where future research, espe-
cially on custard apple, is needed.

PICT,Pune 2 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

2. Literature Survey
Table 2.1: Literature Survey on Fruit Disease Detection
Author Data Dataset Preprocessing Augmentations Model Struc- Performance Outcome
Modal- Size tures Measures
ity
Sneha Ramdas RGB fruit im- 90,380 CNN architec- Not specified CNN, VGG16 Accuracy CNN:
Shegar, Dr. ages ture tailoring, (transfer learn- 98.55%,
Pawan Bhambu VGG16 ing) VGG16:
[2024] [1] 99.05%
Suvarna Pawar, Fruit disease Not men- Feature extrac- Not mentioned Hybrid CNN + Accuracy -
Amruta Surana, images tioned tion ML
et al. [2024] [6]
Bahar Mahmud, RGB images 4,000 (16,000 Batch norm, Scaling, aug- DenseNet Accuracy, F1 99.17%
et al. [2024] [2] w/aug) ReLU mentation Score
Amatullah Ma- Leaf images 87,570 Not mentioned Rotation, zoom, VGG, ResNet Accuracy, F1 85.9%
homodally, et al. shift score
[2024] [9]
Munaf Khalid, RGB images 87,000 Resizing, nor- Not mentioned CNN, Mo- Accuracy CNN:
Oguz Karan malization bileNetV2 89%, Mo-
[2024] [7] bileNetV2:
96%
H.M. Zayani, et RGB image 2,152 Resized to Mosaic (discon- YOLOv8 (53 Precision, Re- 66.67% accu-
al. [2024][8] data 640x640 tinued) layers) call, mAP racy
Feng Xiao, RGB, spec- Not men- Standard pre- Not mentioned YOLO, SSD, Accuracy YOLO:
Haibin Wang, et tral, thermal tioned processing ResNet, Faster Speed,
al.[2023][15] R-CNN ResNet:
Accuracy
Poonam Dhi- Hyperspectral, Not men- Thresholding, Not mentioned SVM, CNN, Classification SVM (best
man, Amandeep RGB tioned color extraction LDA accuracy ML), CNN
Kaur, et al. (best DL)
[2023] [12]
Vasileios Bal- Plant leaf im- Varies by Not mentioned Rotation, flip- CNN, AlexNet, Accuracy, F1 DL methods
afas, et al. ages study ping VGG, ResNet score effective for
[2023] [11] plant disease
detection
Shantilata Not specified N/A Image resizing, Scaling, shadow CNN, VGG, Comparative Provides
Paleia, et al. filtering removal AlexNet analysis insights on
[2023] [10] citrus disease
detection
Rahim Azadnia, JPEG images 3,000 Image process- Rotation, color ResNet-50, Precision, F1 99.63%
et al. [2023] [14] ing manipulation Inception-V3, Score
CNN
Mitali Shewale, Not men- Not men- Resizing, nor- Rotation, flip- CNN Accuracy, F1 99.81%
Rohin Daruwala tioned tioned malization ping, zoom, Score
[2023] [13] contrast
Yousef Algani, Leaf images Not men- Median filter Not mentioned ACO-CNN Accuracy, F1 99.98%
et al. [2023] [16] tioned Score
Aditya Rajbong- Image EXIF 12,000 Not specified Not mentioned CNN, ResNet50, Accuracy, F1 99.37%
shi, et al. [2022] MobileNetV2 score
[17]
Asad Khattak, Citrus 2,293 Keras prepro- Not mentioned CNN Accuracy, F1 95.65%
et al. [2021] [20] fruit/leaf cessing Score
images

PICT,Pune 3 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

3. Requirement Specification and Analysis

The domain of this study is agricultural technology with a focus on plant disease de-
tection, specifically leveraging deep learning models to improve fruit disease diagnosis.
The project aims to provide a systematic review of the application of deep learning al-
gorithms for detecting fruit diseases, particularly targeting the neglected area of custard
apple disease detection.

• Deep Learning Algorithms:

– Explore various deep learning models (CNN, VGG, DenseNet) and their effec-
tiveness in fruit disease detection.

– Analyze performance metrics like accuracy, precision, and recall in the context
of disease classification.

• Fruits Covered:

– Focus on multiple fruits to examine the general trends and advancements.

– Highlight the lack of research on custard apple disease detection to address

this gap.

• Data Considerations:

– Review challenges such as dataset availability, class imbalance, and environ-

mental factors like lighting and noise.

– Investigate data augmentation and preprocessing techniques that can improve

model robustness.

• Practical Application:

– Examine the deployment of these models in real-world settings, including mo-

bile and IoT-based disease monitoring systems.

– Ensure models are lightweight and adaptable for field use by farmers.

• Project Area:

PICT,Pune 4 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

– Computer Vision in Agriculture: The project belongs to the broader

category of computer vision, focusing on its application in detecting plant
diseases from images.

– Machine Learning & Deep Learning: The review will explore how ma-
chine learning and deep learning models contribute to accurate disease detec-
tion in agricultural systems.

3.1 Problem Definition

Fruit diseases significantly impact agricultural productivity and food security. Traditional
methods of disease detection, which rely on manual inspection, are time-consuming, prone
to errors, and often ineffective for early-stage diagnosis. With the growing demand for
efficient and accurate fruit disease detection, the application of deep learning models
offers a promising solution. However, challenges remain in terms of data availability,
environmental variability, and the underrepresentation of certain fruits like custard apple
in existing research. This study aims to address these gaps by exploring the effectiveness
of deep learning models in detecting and classifying fruit diseases, with a special focus
on custard apple.

3.2 Scope
The study encompasses the application of deep learning models in fruit disease detection,
focusing on improving the accuracy and efficiency of diagnosis. It covers the exploration
of various models like CNN, VGG, and DenseNet, along with their performance in de-
tecting diseases across multiple fruits. Special emphasis is placed on addressing the lack
of research for custard apple disease detection. Additionally, the study will review the
challenges related to data availability, environmental variability, and practical deploy-
ment in agricultural settings, including real-world applications like mobile and IoT-based
monitoring systems.

3.3 Objectives
This systematic review aims to evaluate the effectiveness of deep learning models in
detecting fruit diseases, particularly for under-researched fruits like custard apple. It

PICT,Pune 5 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

will identify challenges in implementing these models, explore the underrepresentation

of custard apple in current research, and propose future directions for developing deep
learning-based disease detection systems tailored to this fruit. The goal is to enhance early
disease detection, improve yield quality, and support sustainable agricultural practices.

3.4 Proposed Methodology

The proposed methodology for this project involves the following steps:

• Literature Review Explore various deep learning models (CNN, VGG, DenseNet)
and their effectiveness in fruit disease detection. Analyze performance metrics like
accuracy, precision, and recall in the context of disease classification.

• Dataset Collection Leveraging publicly available datasets from datainbrief.com

streamlines data collection, focusing efforts on preprocessing and organization.
Data augmentation techniques like rotation and scaling will further enrich the
dataset and strengthen model performance.

• Model Selection and Implementation We will implement and compare three

deep learning models—CNN, VGG, and DenseNet. These models will be trained
and tested on the collected dataset to identify the most effective approach for cus-
tard apple disease detection.

• Model Training and Evaluation Each model will be evaluated using perfor-
mance metrics such as accuracy, precision, recall, and F1-score. A confusion matrix
will be used to further analyze model performance. The results will be compared
to determine the best-performing model.

• Preliminary Implementation We will initiate the project’s implementation by

focusing on data preprocessing, basic model setup, and preliminary model evalua-
tion, ensuring that the work is aligned with future real-world deployment for disease
detection in custard apples.

PICT,Pune 6 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

3.5 Project Requirements

3.5.1 Datasets

The Custard Apple Disease Dataset was meticulously collected in the Nimgaon-Bhogi
region of Taluka-Shirur, Pune district, Maharashtra, India. The dataset includes images
from various Custard Apple plantations, reflecting different growth stages, environmental
conditions, and disease manifestations. Key diseases documented include Anthracnose,
Black Canker, Diplodia Rot, Leaf Spot on both fruits and leaves, and Mealy Bug infes-
tations.

• Image Specifications:

– Format: JPG

– Image resolution: 1024 x 768 pixels

3.5.2 Functional Requirements

Our system must posses following capabilities:

• Accurate diseases classification: The system should reliably detect and cate-
gorize various diseases affecting custard apples.

• Real-rime or non-real Detection: To facilitate prompt intervention, the system

should deliver results within a reasonable timeframe.

• Robustness to environmental variations: The system must effectively handle

fluctuations in lighting, background, and other environmental conditions.

• User friendly interface: The system should feature a straightforward and intu-
itive interface that allows users to easily upload images and receive results.

3.5.3 Non Functional Requirements

• Performance: The system should operate efficiently regarding processing time

and the utilization of computational resources.

• Scalability: The system must be capable of accommodating increasing volumes of

data and user demand.

PICT,Pune 7 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

• Usability: The system should be user-friendly, ensuring accessibility for individuals

with limited technical knowledge.

• Reliability: The system should demonstrate consistent and dependable perfor-

mance.

• Security: If deployed in a production environment, the system must incorporate

appropriate security measures to safeguard sensitive information.

3.5.4 Hardware Requirements

• Computer system: A desktop or laptop with at least

– Processor: Intel i5 or equivalent

– RAM: 8 GB or more

– Storage: 500 GB HDD or SSD

• Camera Equipment: High-resolution cameras for image capture, capable of cap-

turing images in varying lighting conditions.

• IoT Devices (optional): Sensors for environmental monitoring, if integrating

real-time disease monitoring.

3.5.5 Software Requirements

• Programming languages: Python for implementing machine learning algorithms.

• Deep learning libraries: TensorFlow or PyTorch for developing deep learning

models.

• Image processing libraries: OpenCV and PIL for image pre-processing and
manipulation.

• Development Environments: IDEs such as PyCharm or Jupyter Notebook for

coding and experimentation.

• Databases: SQLite or PostgreSQL for storing user data and image metadata.

• Web Framework: Flask or Django for developing the user interface.

PICT,Pune 8 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

3.6 Project Plan

3.6.1 Project Resources

• Current Resources:

– Literature: Access to research papers, journals, and articles for the survey.

– Team: Group members contributing to literature review and preliminary test-

ing.

– Hardware: Any systems or cloud platforms for running basic models during
testing phases.

• Future Resources:

– Data: Collection of custard apple disease images.

– High-performance hardware: Potential need for GPUs or cloud platforms like

Google Colab for large-scale model training.

3.6.2 Module Split-up

• Survey Phase(Current work):

– Literature Review: Identifying existing research on fruit disease detection,

focusing on models like CNN, VGG, DenseNet.

– TResearch Gap Analysis: Identifying gaps (especially the lack of research on

custard apple disease detection).

• Preliminary Implementation Phase (20-30%):

– Dataset Collection: Collecting available images of custard apple diseases or

closely related datasets.

– Model Training: Testing CNN, VGG, and DenseNet models on this dataset to
compare performance.

– Evaluation: Comparing accuracy and other metrics to show which model per-
forms best for disease detection.

PICT,Pune 9 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

3.6.3 Project Team Role and Responsibilities

Taks divided among team members:

• Gayatri Bhangle: Leads the literature review, compiles research on models.

• Ashutosh Patil: Responsible for dataset collection and preprocessing.

• Sejal Kadam: Implements the initial model training and evaluation (testing).

• Poonam Vetal: Handles report writing and documentation.

3.6.4 Project Plan 3.0

Outline a timeline for our project, covering both the survey and the initial implementa-
tion:

• Weeks 1-3: Literature review and research gap analysis.

• Weeks 4-5: Preliminary dataset collection and preprocessing.

• Weeks 6-7: Train and evaluate CNN, VGG, DenseNet on the collected dataset.

• Week 8: Finalize the report, documenting both the survey and the initial imple-
mentation results.

3.6.5 PERT Table

Task Optimistic Time (days) Most Likely Time (days) Pessimistic Time (days)
Literature Review 7 10 14
Dataset Collection 5 7 10
Model Implementation 7 10 14
Report Writing 5 7 10

Table 3.1: PERT Table for Project Tasks

PICT,Pune 10 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

4. System Analysis and Design

4.1 System Architecture

The architecture of the proposed Custard Apple Disease Detection system primarily
consists of the following components:

1. User Interface (Web/Mobile App):

• The front-end interface allows users, particularly farmers, to easily interact

with the system. It enables the uploading of images of custard apple fruits
and leaves that may be affected by diseases.

• This interface can be accessed through a web application or a mobile app,

providing a user-friendly experience for individuals without technical expertise.

2. Image Processing Model:

• Once an image is uploaded, the system performs image preprocessing to ensure

the data is in a suitable format for analysis. This includes operations like
resizing, normalization, and enhancing image quality.

• The interface can be accessed through a web application or a mobile app,

providing a user-friendly experience for individuals without technical expertise.

3. Deep Learning Model (Disease Detection):

• The core of the architecture is the deep learning model, typically a Convolu-
tional Neural Network (CNN) trained to recognize and classify various diseases
affecting custard apple plants.

• This model is hosted on the server-side and processes the preprocessed images
to detect diseases such as Anthracnose, Black Canker, Diplodia Rot, Leaf Spot,
etc.

• After classification, the model returns results, including the identified disease
and confidence levels, to the user interface.

PICT,Pune 11 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

4. Data Storage:

• The architecture incorporates a data storage system to store relevant data,

including the uploaded images, preprocessing results, model outputs, and user
information.

• This data is vital for future analysis, model improvement, and generating
reports on disease trends.

5. Client-Server Model:

• The system operates on a client-server architecture, where the client-side (user

interface) communicates with the server-side, which is responsible for the pro-
cessing and execution of deep learning models.

• Images are uploaded from the client to the server, where the deep learning
model processes the data. The results are then sent back to the client in the
form of disease diagnoses and management suggestions.

6. Real-time Monitoring with IoT Devices:

• To support continuous and real-time disease monitoring, IoT devices or mobile

cameras can be deployed in custard apple plantations.

• These devices are integrated into the architecture to automatically capture

images at regular intervals and upload them to the system for automated
detection and monitoring without manual intervention.

7. Backend Processing:

• The back-end processes images and performs disease classification. This part of
the system includes the image preprocessing pipeline, deep learning inference
engine, and storage management.

• It manages all operations needed to analyze the images and send the results
back to the front-end for user interaction.

8. Report Generation and Feedback:

PICT,Pune 12 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

• After detecting a disease, the system generates a detailed report with the
diagnosed disease, suggested treatments, and preventive measures. This report
is displayed on the user interface and can be downloaded for future reference.

Figure 4.1: System Architecture

PICT,Pune 13 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

4.2 Necessary UML Diagrams

4.2.1 Use Case Diagram

Figure 4.2: Use Case Diagram

4.2.2 DFD

Figure 4.3: Data Flow Diagram

PICT,Pune 14 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

4.2.3 Activity Diagram

Figure 4.4: Activity Diagram

PICT,Pune 15 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

4.2.4 Class Diagram

Figure 4.5: Class Diagram

PICT,Pune 16 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

5. Implementation
This section outlines the preliminary stages of the implementation of our project, which

is focused on disease detection in custard apples using deep learning models.

5.1 Stages of Implementation

5.1.1 Data Preprocessing

Data preprocessing is a crucial step in preparing the dataset for deep learning models. For

the custard apple disease detection, the preprocessing will involve collecting image data

of diseased and healthy custard apples. The dataset will then be cleaned and prepared

by performing the following operations:

• Image resizing: All images will be resized to a uniform dimension to ensure

consistency for input into the deep learning models.

• Normalisation: The pixel values of the images will be normalized to fall within a

specified range (e.g., 0 to 1) to ensure that all images are on the same scale.

• Noise Reduction: Any noise or distortions in the images will be filtered out using

image processing techniques to improve the quality of the dataset.

• Data augmentation:Since the dataset might be limited, data augmentation tech-

niques such as rotation, flipping, zooming, and contrast adjustments will be applied

to artificially increase the size of the dataset and introduce variability.

This preprocessing will ensure that the deep learning models receive high-quality, stan-

dardized input data, thereby improving the accuracy of disease detection.

5.1.2 Implementation of Modules

The actual implementation of the deep learning models will take place in the next

PICT,Pune 17 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

semester. For the disease detection system, we plan to experiment with several state-of-

the-art models, including:

• Convolutional Neural Network: CNNs are well-suited for image classification

tasks and will serve as the baseline model for disease detection.

• VGG: As a transfer learning model, VGG16 will be used to leverage pre-trained

weights and adapt them to our dataset for enhanced accuracy.

• DenseNet: DenseNet’s capability to strengthen feature propagation and reduce

the number of parameters will be explored to further improve the model’s perfor-

mance.

5.2 Experimentation Setup

Although the experimentation setup has not yet been finalized, we have a theoretical
framework for how the experiments will be conducted:

• Hardware Requirements: The model training will be performed on a high-

performance GPU setup to efficiently handle the computational demands of deep
learning algorithms. A workstation or cloud-based platforms like Google Colab or
Jupyter notebook will be considered.

• Model training: The implementation will be done using Python programming

language with popular deep learning libraries such as TensorFlow and Pillow. Ad-
ditionally, OpenCV will be used for image processing tasks.

• DenseNet: Each model will undergo training on a portion of the dataset (typically
80% for training and 20% for validation). Techniques such as cross-validation and
hyperparameter tuning will be applied to optimize the model’s performance.

• Evaluation metrics: The performance of the models will be evaluated based on

several metrics, including accuracy, precision, recall, and F1 score. These metrics
will provide insights into how well the model is performing in classifying diseases

PICT,Pune 18 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

in custard apples. Once the preprocessing is complete and the models are imple-
mented, we will move towards fine-tuning the models and preparing for deployment.

PICT,Pune 19 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

6. Conclusion and Future Scope

6.1 Conclusion

There is no doubt that deep learning has revolutionized the machine learning process

in general and created a more efficient way to work on computer systems. This article

provides a comprehensive review of deep learning models for fruit disease detection. This

is a review of all the articles in this field over the last few years. It provides insight

into which model is more suitable and what kind of data is needed to train such a model.

Each study has something new to investigate and the literature survey provides a detailed

description of each paper as well as main results such as the materials and methods used.

There is also a small body of published literature in this area, which is discussed in this

article. Commonly used prioritization, segmentation and deep learning are discussed in

depth. However, more accurate and diverse data as needed to support the research. This

article will be very useful for researchers who plan to study this fruit detection topics and

strengthen the existing model and explore all other new models that can be used with

data.

6.2 Limitations of the Project

Despite the promising potential of our proposed deep learning approach for disease de-

tection in custard apples, several limitations must be acknowledged. First, the reliance

on image data means that the system’s performance may vary significantly based on the

quality and resolution of the images captured. Variability in lighting conditions, camera

settings, and the angle of image capture can impact the accuracy of the model, leading to

potential misclassification of diseases. Additionally, the current study primarily focuses

on a limited dataset, which may not encompass the full spectrum of diseases affecting

custard apples. Consequently, the model’s generalizability could be constrained. Further-

PICT,Pune 20 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

more, the computational resources required for training and implementing deep learning

models can be substantial, making it less accessible for small-scale farmers or stakeholders

with limited technological infrastructure. Finally, while the project emphasizes accuracy,

it does not extensively address the interpretability of the model’s predictions, which is

crucial for practical deployment in agricultural settings.

6.3 Future Scope

Looking ahead, there are several avenues for further exploration and enhancement of our

project. Future work could focus on expanding the dataset to include a wider variety of

custard apple diseases and more diverse environmental conditions, thereby improving the

model’s robustness and accuracy. Integrating advanced image preprocessing techniques

and data augmentation methods could also help in mitigating issues related to image qual-

ity and variability. Additionally, the development of a mobile application could provide

users with real-time disease detection capabilities, facilitating quicker intervention and

management practices. Implementing transfer learning with more complex architectures

or ensembling techniques may yield even better performance. Moreover, future research

could explore the interopretability of deep learning models to make them more user-

friendly for farmers, allowing them to understand the rationale behind specific disease

classifications. Collaborations with agricultural experts could also enrich the project by

providing domain knowledge that informs model adjustments and improvements. Over-

all, the future scope of this project is broad, with numerous possibilities for enhancing

its practical application in agriculture.

PICT,Pune 21 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

Bibliography

[1] Sneha Ramdas Shegar, Dr. Pawan Bhambu, “Fruit Quality Analysis and Disease
Detection using Deep Learning Techniques”, vol. 32, J. Electrical Systems 20-3s
(2024): 755-762

[2] Mahmud, B. U., Al Mamun, A., Hossen, M. J., Hong, G. Y., & Jahan, B. (2024).
Light-weight deep learning model for accelerating the classification of mango-leaf dis-
ease. Emerging science journal, 8(1), 28-42.

[3] Madeira, M., Porfı́rio, R. P., Santos, P. A., & Madeira, R. N. (2024). AI-powered
Solution for Plant Disease Detection in Viticulture,Procedia Computer Science, 238,
468-475.

[4] Varadraj Jadhav, Harshal Giri, Harish Thore, Kshitij Patil, Hemant Pawar,
Dr. Tushar Phadtare, “Multiple Fruit Disease Detection Using Deep Learning”,
IJFMR240321378 Volume 6, Issue 3, May-June 2024

[5] Thite, S., Patil, K., Jadhav, R., Suryawanshi, Y., & Chumchu, P. (2024). Empowering
agricultural research: A comprehensive custard apple (Annona squamosa) disease
dataset for precise detection. Data in Brief, 53, 110078.

[6] Dr. Suvarna Eknath Pawar, Dr. Amruta V Surana, Dr. Pooja Sharma, Dr. Ramachan-
dra Pujeri,”(2023). Fruit Disease Detection and Classification using Machine Learning
and Deep Learning Techniques International Journal of Intelligent Systems and Ap-
plications in Engineering, 12(4s), 440–453.

[7] Khalid, M. M., & Karan, O. (2023). Deep Learning for Plant Disease Detec-
tionInternational Journal of Mathematics, Statistics, and Computer Science, 2,
75–84.

[8] Zayani, H.M., Ammar, I., Ghodhbani, R., Maqbool, A., Saidani, T., Slimane, J.B.,
Kachoukh, A., Kouki, M., Kallel, M., Alsuwaylimi, A.A. and Alenezi, S.M. 2024.
Deep Learning for Tomato Disease Detection with YOLOv8. Engineering, Technology
& Applied Science Research. ,14, 2 (Apr. 2024), 13584–13591.

PICT,Pune 22 Dept. of Information Technology

 Custard Apple Disease Prediction with Deep Learning

[9] Mahomodally, A. F. H., Suddul, G., & Armoogum, S. (2023). Machine learning tech-
niques for plant disease detection: an evaluation with a customized dataset. Int J Inf
& Commun Technol, 12(2), 127-139.

[10] Palei, Shantilata, Santi Kumari Behera, and Prabira Kumar Sethy. ”A systematic
review of citrus disease perceptions and fruit grading using machine vision.” Procedia
Computer Science 218 (2023): 2504-2519..

[11] Balafas, V., Karantoumanis, E., Louta, M., & Ploskas, N. (2023). Machine learning
and deep learning for plant disease classification and detection.IEEE Access.

[12] Dhiman, P., Kaur, A., Balasaraswathi, V. R., Gulzar, Y., Alwan, A. A., & Hamid,
Y. (2023). Image acquisition, preprocessing and classification of citrus fruit diseases:
A systematic literature review . Sustainability, 15(12), 9643.

[13] Shewale, Mitali V., and Rohin D. Daruwala. ”High performance deep learning ar-
chitecture for early detection and classification of plant leaf disease. Journal of Agri-
culture and Food Research14 (2023): 100675.

[14] Azadnia, Rahim, Saman Fouladi, and Ahmad Jahanbakhshi. ”Intelligent detection
and waste control of hawthorn fruit based on ripening level using machine vision
system and deep learning techniques.” Results in Engineering 17 (2023): 100891.

[15] Xiao, F., Wang, H., Xu, Y., & Zhang, R. (2023). Fruit detection and recognition
based on deep learning for automatic harvesting: An overview and review.Agronomy,
13(6), 1625.

[16] Rajbongshi, A., Sazzad, S., Shakil, R., Akter, B., & Sara, U. (2022). A comprehen-
sive guava leaves and fruits dataset for guava disease recognition. Data in Brief, 42,
108174.

[17] Andrew J., Jennifer Eunice R, Daniela Elena Popescu, M. Kalpana Chowdary and
Jude Hemanth D, “Deep Learning-Based Leaf Disease Detection in Crops Using Im-
ages for Agricultural Applications”, Agronomy 2022, 12, 2395

[18] Mr. Thangavel. M -AP/ECE, Gayathri P K, Sabari K R, Prathiksha V,” Plant

Leaf Disease Detection using Deep Learning”, IInternational Journal of Engineering
Research & Technology (IJERT), Volume 10, Issue 08, Special Issue - 2022.

[19] Khattak, A., Asghar, M. U., Batool, U., Asghar, M. Z., Ullah, H., Al-Rakhami, M.,
& Gumaei, A. (2021). Automatic detection of citrus fruit and leaves diseases using
deep neural network model. IEEE access, 9, 112942-112954.

PICT,Pune 23 Dept. of Information Technology

Sample Document
Sample Document
Sample Document
Sample Document