Optimization of Convolutional Neural Network (CNN) Using Transfer Learning for Disease Identification in Rice Leaf Images

Abstract

Rice productivity, as one of the key commodities in Southeast Asia, is often hindered by various plant diseases such as Rice Blast, Bacterial Leaf Blight, and Brown Spot, which can cause significant economic losses for farmers. This study aims to develop an automated rice leaf disease detection system using deep learning, specifically leveraging the Convolutional Neural Networks (CNN) architecture with a transfer learning approach. The dataset used comprises 10,407 images of rice leaves categorized into 10 classes, including various diseases and healthy leaves. The dataset is divided into three parts: 80% (8,323 images) for training, 15% (1,557 images) for validation, and 5% (527 images) for testing. The trained EfficientNetB0 model was utilized for feature extraction and classification. The evaluation used accuracy, precision, recall, and F1-score metrics based on a confusion matrix. The results revealed that the model achieved a global accuracy of 98.86%, a micro precision of 100%, a micro recall of 99.42%, and a micro F1-score of 99.70%. These findings underscore the effectiveness of the proposed approach in automating rice leaf disease detection, providing a significant contribution to technology-based agricultural solutions.