ResViT-Rice: A Deep Learning Model Combining Residual Module and Transformer Encoder for Accurate Detection of Rice Diseases

Rice is a staple food for over half of the global population, but it faces significant yield losses: up to 52% due leaf blast disease and brown spot diseases, respectively. This study aimed at proposing hybrid architecture, namely ResViT-Rice, by taking advantage both CNN transformer accurate detection diseases. We employed ResNet as backbone network establish model introduced encoder component from architecture. The convolutional block attention module was also integrated ResViT-Rice further enhance feature-extraction ability. processed 1648 training 104 testing images two diseases healthy class. To verify effectiveness proposed we conducted comparative evaluation with popular deep learning models. experimental result suggested that achieved promising results in rice disease-detection task, highest accuracy reaching 0.9904. corresponding precision, recall, F1-score were all 0.96, an AUC 0.9987, loss rate 0.0042. In conclusion, can better extract features different thereby providing more robust classification output.