基于CLD-YOLO的高粱病虫害识别方法OA

To address the challenges of sorghum pest and disease detection in natural environments,we propose a sorghum pest and disease detection model,CLD-YOLO,based on an improved YOLOv8n.First,the ODConv structure is introduced into the backbone network to replace traditional convolutions.This modification reduces background noise interference and enhances the model's focus on sorghum leaf features,thereby improving detection accuracy.Additionally,the LSKA attention mechanism is incorporated after the SPPF module in the backbone network,effectively combining the strengths of convolution and self-attention to highlight the detailed features of diseases and pests,further enhancing detection precision.Finally,the DySample dynamic upsampling technique is applied in the neck network to enable more precise localization and identification of sorghum pest and disease-affected areas,significantly boosting detection performance.Experimental results show that CLD-YOLO achieves a mAP0.5 of 90.9%and an accuracy of 90.5%in sorghum pest and disease image recognition tasks,with a computational load of only 7.3 GFLOPs.Compared to YOLOv8n,mAP0.5 improved by 3.8 percentage points,accuracy by 5.1 percentage points,and computational load decreased by 9.9%.The improved model offers higher recognition accuracy and more precise target localization,making it suitable for accurate detection of sorghum pests and diseases in natural environments.It provides a theoretical basis for the intelligent recognition of sorghum pests and diseases.