Strip Steel Surface Defect Detection Algorithm Based on Lightweight YOLOv5s
The detection of defects on the surface of strip steel is very important for industrial production.Aiming at the problems of large number of parameters and large computation volume of strip steel sur-face defect detection algorithms in industrial scenarios,a lightweight GDSB-YOLOv5 algorithm is pro-posed for detecting defects on strip steel.Firstly,GhostNet is used as the backbone network to reduce the number of parameters and the computation amount;meanwhile,depth-separable convolution(DSConv)is introduced to further reduce the number of parameters;secondly,Squeeze-and-Excitation(SE)attention module is added after the three concatenation to effectively improve the extraction of tar-get features.Finally,the BiFPN weighted bidirectional pyramid structure is introduced to replace the FPN+PAN structure to improve the fusion efficiency of the feature map.The experimental results show that the average detection accuracy of GDSB-YOLOv5 is 78％,which is 1.3％higher than the original YOLOv5s algorithm,and the computational and parametric quantities are reduced by 51.2％and 26％,respectively.The detection algorithm achieves the lightweight of the algorithm while ensuring the aver-age detection accuracy.
国家科技期刊平台
NSTL
万方数据
