基于改进YOLOv8n的钢材表面缺陷检测算法OA

In order to solve the problems of insufficient feature extraction ability and low detection accura-cy of complex defects such as irregular shape and significant scale variation,a steel surface defect detec-tion algorithm YOLOv8n-OMD based on improved YOLOv8n was proposed.Firstly,in the backbone net-work,the Online Convolutional Re-parameterization(OREPA)is used to reparameterize the complex network structure layer into a single convolutional layer,which has a low computational cost while main-taining the feature extraction ability.Secondly,the Median-Enhanced Channel and Spatial Attention(MECS)was designed and integrated to the end of the backbone network to enhance the extraction of im-portant features and improve the detection accuracy.Finally,the C2f_DCNv4 module was constructed by combining the deformable convolution DCNv4 and introduced into the neck network to increase sampling in the effective receptive field and effective position,thereby capturing the details of complex shape fea-tures more accurately.Experimental results show that the YOLOv8n-OMD algorithm achieves 82.2%mAP@0.5,48.8%mAP@0.5∶0.95,with a computational cost of 7.1×109 FLOPs on the NEU-DET steel surface defect dataset.Compared with the benchmark algorithm YOLOv8n,the mAP@0.5 and mAP@0.5∶0.95 are increased by 3.3%and 1.5%,respectively,and the computational cost is re-duced by 12.3%,which proves the effectiveness and practicability of the YOLOv8n-OMD algorithm for steel surface defect detection.