Lightweighted Cotton Pest and Disease Detection Method Based on Improving YOLOv8n

Xiong Bin; Wei Dengfeng; Zhang Jianmin

doi:10.14083/j.issn.1001-4942.2025.11.020

Abstract：In order to solve the problems of complex and diverse background of cotton disease and pest detection in actual agricultural scenarios,the large number of parameters and computational amount of detec-tion models,and the difficulty to deploy in mobile device or embedded devices,a lightweighted cotton disease and pest detection algorithm was proposed on the basis of improving YOLOv8n.Firstly,the DyHead structure was used to replace the original YOLOv8n detection head structure,which could significantly improve the tar-get detection ability of the network.Then,PIoUv2 was introduced as the loss function of the bounding box,which could accelerate the convergence of the boundary loss box and enhance the focusing ability of the medi-um-mass anchor frame.Finally,the Layer Adaptive Amplitude Based Pruning(LAMP)method was used to reduce the detection network,in order to reduce the number of parameters and computational amount of the detection model,and compress the weight file size of the model.The results showed that the improved algo-rithm reduced the number of parameters and computational amount by 57.14%and 61.73%respectively com-pared with the original YOLOv8n;the weight file size of the model was reduced to 2.6 MB,which was 55.93%lower than that of the original YOLOv8n;the mean average accuracy(mAP@0.5)reached 89.5%,which was one percentage point higher than that of the original YOLOv8n.In summary,the algorithm had achieved signif-icant performance improvement in cotton pest detection,and it was more suitable for deployment in embedded devices or mobile terminals through lightweighting the network,which provided a more feasible solution for cotton pest detection in real agricultural scenarios.