Citrus Leaf Disease Detection Using GAN-Based Data Augmentation And Convolutional Neural Networks

Read full text Summarise with VIRAT
Authors: Shubham Madane, Vikul J. Pawar, Kailash D. Kharat
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Citrus crops are one of the pillars of world agri- culture but, due to the presence of pathogenic diseases, the quality of the yield and the volume of production are significantly reduced. The traditional diagnostic methods that used visual examination by skilled agronomists are simply slow, subjective and operationally unfeasible at farm level. The automated deep- learning model presented in the paper aims to classify citrus leaf disease images based on a real-farm dataset of 934 images of 6 types (leaf blast (115), healthy (167), canker (163), greening (204), black spot (171), and melanose (114)). Agricultural deep learning applications are characterized by two major challenges, namely, a lack of annotated training data and severe class imbalance. Their combination resulted in disease-specific Deep Convolutional Generative Adversarial Networks (DCGANs) to create realistic diseased-leaf examples, in which 500 balanced training examples per minority class are used. A Convolutional Neural Network based on the EfficientNetB0 architecture was then fine-tuned on the synthetically enriched dataset using transfer learning. When only the original, unaugmented farm images are evaluated, it is shown that the GAN-augmented model shows a statistically sig- nificant 92.78% classification accuracy, as compared to 86.42% of the baseline achieved without synthetic augmentation. Precision and recall of the proposed method by class also confirm the strength of the offered solution regarding all types of diseases. The results provide generalizable, scalable approach to deal with data scarcity in agricultural diagnostics and provide a solution that can be implemented in the field to detect citrus diseases at an early stage.

Fields of Study & Subjects

Smart Agriculture and AI Advanced Neural Network Applications Industrial Vision Systems and Defect Detection

Publication details

Journal / source
International Journal of Advances in Signal and Image Sciences
Volume
12
Issue
3s
Pages
1454–1472
Publication date
9 March 2026
Language
EN