基于改进 YOLOv10n 的植物叶片病害轻量化检测模型OA

To address the problems of low efficiency in manual identification and insufficient intelligent monitoring methods for plant leaf diseases,Solanum lycopersicum was used as the experimental material,and a lightweight leaf disease detection algorithm DHS-YOLO based on the improved YOLOv10n network was proposed.First,the DCNv4 deformable convolution was introduced to enhance the model's feature perception ability and inference efficiency for irregular targets of leaf disea-ses.Second,the hierarchical hybrid attention module(C2f_HHA)was adopted to replace the original C2f structure,achieving accurate extraction of key disease features without increasing computational complexity.Finally,a novel spatial-channel-position integrated fusion module(SCPI)was designed in the neck network to strengthen the representation ability of multi-scale feature fusion for leaf diseases.To improve the generalization ability of the model,data augmentation opera-tions including flipping,mosaic,brightness adjustment and Gaussian noise addition were performed on the dataset.S.ly-copersicum leaf disease dataset containing 10 047 images was finally constructed and divided into training,validation and test sets with a ratio of 8:1:1.Experimental results showed that the mAP50 and mAP50-95 values of the DHS-YOLO model for S.lycopersicum leaf disease detection reached 94.7%and 79.9%,respectively,which were 4.1 and 4.0 per-centage points higher than those of the original YOLOv10n model.Meanwhile,the model size was reduced by 6.5%,and both precision and recall were synchronously optimized.