<![CDATA[Rice leaf diseases significantly reduce agricultural productivity, making early and accurate detection essential, particularly in rice-producing regions such as Indonesia. This study proposes an automated rice leaf disease detection system based on Convolutional Neural Networks (CNN) and transfer learning. The dataset, obtained from the Mendeley Data Repository, consists of 6,889 images classified into eight categories: Bacterial Leaf Blight, Brown Spot, Healthy Rice Leaf, Leaf Blast, Leaf Scald, Narrow Brown Leaf Spot, Rice Hispa, and Sheath Blight. The dataset was divided into 70% training, 15% validation, and 15% testing. A baseline CNN model and three pre-trained models—EfficientNetB0, MobileNetV2, and ResNet50—were evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. The baseline CNN achieved a test accuracy of 48.26%, while EfficientNetB0 achieved 58.41%. In contrast, MobileNetV2 and ResNet50 demonstrated significantly better performance, with test accuracies of 79.98% and 76.60%, respectively. MobileNetV2 exhibited the most balanced performance across all classes, showing superior generalization capability and computational efficiency. The best-performing model was integrated into a Streamlit-based application, enabling real-time rice leaf disease detection through image upload. The results confirm that transfer learning substantially improves classification accuracy and robustness compared to conventional CNNs. This study highlights the potential of lightweight deep learning models for practical implementation in smart agriculture systems and provides a reliable solution for automated rice disease detection in real-world conditions.]]>
Copyrights © 2025
