Aiming at the problems of difficult disease spot recognition and slow detection speed caused by complex backgrounds in rice images, this paper proposes an improved YOLOv10 target detection algorithm for automatic identification of rice diseases. Firstly, the backbone network of YOLOv10 is improved by introducing Large Separable Kernel Attention (LSKA) to replace the original Polarized Self-Attention (PSA), which improves the sensitivity to features, enhances the generalization ability and efficiency of the network. In addition, the loss function is replaced by the Normalized Wasserstein Distance (NWD) loss function, which re‐ duces the feature map while preserving more information as much as possible, improving the processing efficiency. Experiments con‐ ducted on the rice dataset show that the improved YOLOv10 algorithm achieves 97.95% on the mAP50 indicator, representing a 1.08% increase compared to the original YOLOv10; and it achieves 78.74% on the mAP50-95 indicator, representing a 1.52% in‐ crease compared to the original YOLOv10.