Comparative Analysis of the Effectiveness of Transformer and Convolutional Neural Network Architectures for Automatic Classification of Rice Leaf Diseases

This article presents a comparative analysis of modern neural network architectures, convolutional neural networks (CNNs) and transformers, for the automatic diagnosis of rice leaf diseases. In the experiments, DenseNet121, ResNet, Vision Transformer (ViT), and MaxViT models were trained and tested, followed by their evaluation in terms of accuracy and computational efficiency. The study was conducted on a large-scale dataset containing real images of healthy and diseased rice leaves, which makes the results highly relevant for agricultural science and practice. The experiments included hyperparameter optimization, application of data augmentation techniques, and the use of loss functions and regularization methods to improve the generalization ability of the models. The evaluation metrics comprised classification accuracy, F1-score, as well as computational efficiency indicators such as prediction time and resource consumption. The results showed that transformer-based models, particularly MaxViT, achieve accuracy of up to 94.10%. This is attributed to their ability to effectively capture both local and global image features through attention mechanisms and deep contextualization. At the same time, CNN architectures such as DenseNet121 and ResNet demonstrate high processing speed and robustness under limited computational resources.