Steel Surface Defect Detection Based on Improved YOLOv4
Aiming at the problems of low detection accuracy of steel surface defects,easy missed detection,false detection,and inaccurate positioning,this paper proposes a steel surface defect detection algorithm based on improved YOLOv4.The excellent anchor frame improves the positioning accuracy and reduces the network loss;Secondly,on the basis of the original feature layer of the YOLOv4 network,a shallow feature layer,that is,a new feature layer with a scale of 104×104,is added to increase the feature detection scale and improve the target of small defects.Detection accuracy;Finally,an attention mechanism is introduced on the basis of the original backbone network,so that the net-work pays more attention to useful information,thereby making the detection more accurate.Comparing the algorithm in this paper with other algorithms on the NEU-DET dataset,the average detection accuracy of the algorithm in this paper has increased by 4.69%to 78.10%compared with the original YOLOv4,which is also better than other current mainstream target detection algorithms.
NETL
NSTL
万方数据
