A Comparative Analysis of Transfer Learning, LeafNet, and Modified LeafNet Models for Precise Classification of Rice Leaf Diseases

Abstract

This research focuses on the timely identification of plant diseases, crucial for efficient crop disease management to mitigate yield loss. The study introduces a methodology for classifying diseases in rice leaves utilizing four distinct deep learning models and a dataset comprising 2658 images of both healthy and diseased rice leaves. The compared models include LeafNet, Modified LeafNet, MobileNetV2, and Xception. Modifications to LeafNet's architectural parameters were implemented in the Modified LeafNet model, while transfer learning techniques were applied to pretrained MobileNetV2 and Xception models. Optimal training hyperparameters were determined by considering various factors such as batch size, data augmentation, learning rate, and optimizers. Notably, the Modified LeafNet model demonstrated the highest accuracies, achieving 97.44% and 87.76% for the validation and testing datasets, respectively. Comparatively, LeafNet achieved 88.92% and 71.84%, Xception achieved 88.64% and 71.95%, and MobileNetV2 achieved 82.10% and 67.68% for validation and test accuracies on the same datasets, respectively. This research significantly contributes to the advancement of automated disease classification systems for rice leaves, thereby enhancing agricultural productivity and sustainability.