The 16th International Conference on Computing,
Communication and Networking Technologies [ICCCNT 2025]
An Approach to Enhance the Agriculture sustainability
by Deep Learning for Banana Leaf Disease Detection
PAPER ID : 2156
COLLEGE NAME : KONGU ENGINEERING COLLEGE
PRESENTED BY : Vanaja T, Ramyasri M M, Yoga M, Pradhanya R S, Sesili A, Jayadhan S
DATE : 09-07-2025
LIST OF CONTENT
★ ABSTRACT
★ INTRODUCTION
★ LITERATURE SURVEY
★ PROBLEM STATEMENT
★ DATASET DESCRIPTION
★ PROPOSED WORK
★ METHODOLOGY
★ DATA PREPROCESSING
★ ALGORITHMS APPLIED
★ KEY FINDINGS AND FUTURE WORK
★ REFERENCES
ABSTRACT
❖ Banana is a vital global crop, but its yield and quality are threatened by leaf diseases,
making early detection and accurate classification essential for effective management.
❖ This study proposes an AI-based approach using machine learning and deep learning
technique CNN and ResNet50 to identify four banana leaf conditions: Fusarium Wilt,
Black Sigatoka, Cordana Leaf Spot, and healthy leaves.
❖ A real-world dataset from farms in Tamil Nadu was used. Models were evaluated on
accuracy, precision, and recall.
❖ ResNet50 highlighted the effectiveness of deep learning in real-time crop disease
detection and promoting sustainable agriculture.
INTRODUCTION
❖ This study focuses on the early detection and classification of banana leaf diseases,
a major challenge affecting crop yield and quality globally. Leaf diseases like
Fusarium Wilt, Black Sigatoka, and Cordana Leaf Spot severely impact
productivity, requiring timely identification for effective intervention.
❖ Real-world banana leaf images collected from farms in Tiruchengode, Tamil Nadu
are used to ensure realistic and diverse data. Comprehensive preprocessing and
augmentation techniques are applied to enhance model accuracy and robustness.
❖ By combining traditional machine learning (SVM, RF, NB, DT) with advanced deep
learning models (CNN, ResNet50), the research evaluates multiple predictive
frameworks. This AI-driven approach supports precision agriculture, enabling
farmers to make early, informed decisions and improve crop health management.
LITERATURE SURVEY
PROBLEM STATEMENT
❖ Complexity in Early Detection: Banana plants are highly vulnerable to various leaf
diseases that significantly reduce yield and crop quality. Early and accurate detection is
critical but difficult due to subtle visual symptoms.
❖ Delayed and Inconsistent Identification:Traditional methods rely on manual
inspection, which is time-consuming, subjective, and often fails to detect diseases at
early stages, especially in large-scale farms.
❖ Lack of Technological Integration in Farming: Many agricultural regions lack access
to automated disease monitoring systems, limiting farmers' ability to respond quickly
and accurately to outbreaks.
❖ Objective of the Study: This study aims to apply machine learning and deep learning
techniques—including CNN and ResNet50—on real-time image data to develop a
reliable, automated system for the early detection and classification of banana leaf
diseases.
DATASET DESCRIPTION
❖ Dataset Source & Type: 455 high-resolution banana leaf images (10–12 MB) collected from real-
time farm environments in Tiruchengode, Tamil Nadu.
❖ Disease Classes:
➢ Fusarium Wilt
➢ Black Sigatoka
➢ Cordana Leaf Spot
➢ Healthy Leaves
❖ Train-Test Split:
➢ Training: 364 images (80%)
➢ Testing: 91 images (20%)
❖ Image Preprocessing: Includes resizing, grayscale conversion (for ML), and augmentation
techniques like rotation, flipping, zooming, and brightness adjustment.
❖ Application: Used for training machine learning (SVM, RF, NB, DT) and deep learning models
(CNN, ResNet50) for multi-class disease classification.
PROPOSED WORK
METHODOLOGY
❖ Python-based implementation utilizing machine learning libraries for creating
predictive models and image processing capabilities.
❖ Real-time dataset sourced from banana farms in Tiruchengode, Tamil Nadu,
ensuring authentic agricultural conditions and diverse disease presentations.
❖ Integration of traditional machine learning algorithms (SVM, Random Forest,
Naive Bayes, Decision Tree) with deep learning architectures (CNN, ResNet50).
❖ Models assessed using accuracy, precision, recall, and F1-score metrics with
80:20 train-test split for robust performance validation.
❖ ResNet50 pre-trained on ImageNet dataset with fine-tuning for banana leaf
disease classification tasks.
DATA PREPROCESSING
❖ Raw high-resolution images resized to
maintain consistent input dimensions across all
models.
❖ Advanced techniques including rotation,
flipping, zooming, and brightness adjustment
to artificially increase dataset diversity and
improve model robustness.
❖ Grayscale conversion for traditional ML
models, HOG feature extraction, standard
scaling for normalization, and 80:20 train-test
split.
ALGORITHMS APPLIED
EVALUATION OF THE APPLIED ALGORITHMS
KEY FINDINGS AND FUTURE WORK
Key Findings:
❖ ResNet50 and CNN models significantly outperformed traditional machine
learning, with ResNet50 achieving 93% accuracy in classifying banana leaf
diseases.
❖ The system works well with real farming data, making it suitable for immediate
agricultural use and enabling early disease identification to support precision
agriculture.
Future Work:
❖ The plan is to optimize models for mobile and edge devices in farming
environments and integrate sensor data and environmental factors for better
accuracy.
❖ Future efforts will focus on developing systems that can continuously learn
from new data and changing agricultural conditions.
REFERENCES
1. Ferentinos, K.P. (2018). Deep learning models for plant disease detection and diagnosis.
Computers and Electronics in Agriculture, 145, 311–318.
2. Banerjee, D., et al. (2023). Precision Agriculture: Classifying Banana Leaf Diseases with Hybrid
Deep Learning Models. IEEE 8th International Conference for Convergence in Technology
(I2CT).
3. Saranya, N., et al. (2020). Detection of banana leaf and fruit diseases using neural networks.
IEEE 2nd International Conference on Inventive Research in Computing Applications (ICIRCA).
4. Sangeetha, T., et al. (2020). Detection of pest and disease in banana leaf using convolution
Random Forest. Test Engineering and Management, 83, 3727–3735.
5. Deenan, S., et al. (2020). Image segmentation algorithms for Banana leaf disease diagnosis.
Journal of The Institution of Engineers (India): Series C, 101, 807–820.
