Steel Surface Defect Detection Based on YOLOv5-GCE
Aiming at the problem of low accuracy,easy omission and false detection in traditional small-size defect detection,an improved YOLOv5-GCE method is proposed for detecting surface defects of steel.Firstly,the GhostNet modular is exploited in YOLOv5′s backbone network to replace the residual modular in CSP1 in order to reduce the amount of modelar factors and mathematical complexity.Secondly,the CA attention mechanism is introduced to enable the network to prioritize the key features relating to small tar-gets,thus enhancing its feature extraction and positioning capabilities and improving the accuracy of small target detection.Finally,the traditional GIoU loss function is replaced by EIoU loss function,which im-proves the convergence speed and regression precision of the model.The experimental results show that the mAP value of the YOLOV5-GCE algorithm on the NEU-DET dataset is 81.4%,which is 4.5%higher than that of the original YOLOv5 algorithm,and the detection speed reaches 40 fps.Moreover,the model size of the algorithm is small which can be applied to the application scenarios of mobile target detection.
defect detectionYOLOv5GhostNet moduleattention mechanismloss function
NETL
NSTL
万方数据
