<![CDATA[This study compares four transfer learning-based CNN models, namely VGG19, ResNet152, MobileNetV2, and DenseNet121, for the classification of 10 classes of rice leaf diseases. Evaluation results on the test dataset show that ResNet152 achieves the best performance with an accuracy of 0.9553, precision of 0.9589, recall of 0.9553, and F1-score of 0.9558, followed by DenseNet121 (accuracy 0.9433), MobileNetV2 (0.9353), and VGG19 (0.9247). ResNet152 excels in recognizing complex features through its skip connection mechanism, while DenseNet121 is more efficient with the lowest validation loss. MobileNetV2 is the lightest and fastest model, making it suitable for resource-limited devices. Based on the confusion matrix analysis, all models are able to classify the neck blast class perfectly; however, misclassifications still occur among visually similar classes such as brown spot, narrow brown spot, and leaf blast. Overall, transfer learning is proven effective for rice leaf disease classification, with ResNet152 and DenseNet121 being the most recommended models.]]>
Copyrights © 2026
