Augmentation strategies for plant disease classification

Chia, Wan Jun (2025) Augmentation strategies for plant disease classification. Final Year Project, UTAR.

Preview

PDF Download (3103Kb) | Preview

Abstract

This study explores the effectiveness of various data augmentation strategies for enhancing plant disease classification using the LeafGAN model. We propose a novel approach that integrates leaf region and disease symptom masking to improve the quality of synthetic images and, consequently, the performance of plant disease models. Three different configurations of the LeafGAN model were tested, with each model applying distinct masking techniques: LeafGAN with LFLSeg uses basic LeafGAN outputs, SingleMask-LeafGAN applies leaf region masking to isolate the leaf from the background, and DualMask-LeafGAN combines both leaf region and disease symptom masking for enhanced disease simulation. The models were evaluated based on their ability to generate realistic disease progression and recovery images, which were then used for data augmentation. Results show that DualMask-LeafGAN, incorporating both masking strategies, produced the most realistic and high-fidelity images, leading to superior augmentation quality. These findings highlight the potential of advanced data augmentation strategies in improving plant disease simulation, emphasizing the importance of targeted feature masking in enhancing the generalization and robustness of disease classification models in agricultural applications. Area of Study: Deep Learning for Image Segmentation and Classification, Computer Vision in Agriculture Keywords: Data Augmentation, Deep Learning, LeafGAN, Image Masking, Plant Disease Detection

Actions (login required)