Comparative Analysis of Convolutional Neural Network Architectures in Classifying Types of Rice Leaf Diseases

Abstract

Rice diseases are a serious threat to rice production in Indonesia, with annual losses reaching 200,000-300,000 tons. Early detection and accurate diagnosis for rice leaf diseases are essential for effective control, but manual methods require a lot of time and effort. This study aims to compare the performance of Convolutional Neural Network (CNN) architectures, namely Xception and NASNetMobile, in classifying rice leaf disease types. The methods used include collecting a dataset of 670 rice leaf images, data preprocessing, CNN model design and training, and performance evaluation using accuracy, precision, recall, and F1-score metrics. The results show that the Xception architecture outperforms NASNetMobile with 93% accuracy versus 83%. Xception also showed more stable and consistent performance in classifying different types of rice leaf diseases, especially for Bacterial leaf blight and Brown spots. This study provides new insights into the effectiveness of CNN architecture in plant disease classification, which can be beneficial for the development of more accurate and efficient disease detection systems in the future.