<![CDATA[Rice sustains over half of the global population; however, leaf diseases such as blight, brown spot, and leaf smut significantly reduce crop yields. Early and accurate detection is essential for supporting sustainable agriculture practices. This study proposes a ResNet-50-based deep learning model for rice leaf diseases classification using a dataset of 1,150 field images collected from Sleman, Indonesia, which was expanded to 2,000 images through augmentation techniques, including rotation, flipping, zooming, and brightness adjustment. Model performance was evaluated using both hold-out validation and 10-fold cross-validation with accuracy, precision, recall, and F1-Score metrics. The application of data augmentation improved hold-out validation accuracy from 82.4% to 88.1%. Meanwhile, 10-fold-cross-validation yielded a substantially higher average accuracy of 99.6%. This discrepancy suggests potential sensitivity to data partitioning and indicates the need for careful interpretation, as cross-validation may introduce optimistic estimates under certain conditions. Although the proposed approach demonstrates strong performance in distinguishing visually similar diseases, this study limited by the use of a single-region dataset, which may affect generalizability. Therefore, the integration of ResNet-50, augmentation, and cross-validation shows promising results for early disease detection, while further validation on more diverse datasets is required to support it application in real-world precision agriculture systems.]]>
