This research introduces a method for optimizing deep learning models designed to identify diseases in crop images, such as tungro, blast, and blight. By utilizing Keras Tuner with the Random Search technique, we fine-tune the hyperparameters of a Convolutional Neural Network (CNN) to improve its performance in disease classification. The optimized model achieves an impressive test accuracy of 97.92% and a validation accuracy of 98.61%, demonstrating its effectiveness. These results underscore the model’s potential for use in smart agriculture, offering a valuable tool for early detection of rice crop diseases.