<![CDATA[Rice plant diseases significantly affect crop productivity and require accurate and timely identification to support effective management. This study proposes a rice leaf disease classification approach using the ResNet-50 convolutional neural network and compares the performance of three optimization algorithms, namely ADAM, Stochastic Gradient Descent (SGD), and RMSProp. The model was trained and evaluated on a rice leaf image dataset consisting of four classes BrownSpot, Healthy, Hispa, and LeafBlast. The dataset contains visual variations in color, texture, and disease patterns that influence classification performance. Performance was assessed using training accuracy, loss, precision, recall, F1-score, and confusion matrix analysis. These evaluation metrics provide a comprehensive measurement of model effectiveness and class-wise prediction behavior. Experimental results show that the ADAM optimizer achieved the best performance with a training accuracy of 75.84%, followed by RMSProp at 74.60%, while SGD obtained the lowest accuracy of 71.34%. The differences in performance highlight the impact of optimization strategies on deep neural network training stability. Class-wise evaluation indicates that the model performed well in detecting BrownSpot and Healthy classes, but showed lower performance on the Hispa class across all optimizers. This limitation is influenced by the visual similarity of Hispa symptoms to other classes. These findings demonstrate that adaptive learning rate–based optimizers provide faster convergence and better classification performance for deep learning–based rice disease detection. The results support the use of optimized convolutional neural networks for image-based agricultural applications.]]>
