Steel Surface Defect Detection Based On Improved YOLOv5

Steel is an indispensable material for national construction and realization of the four modernizations,and has an important application in all walks of life.Due to problems such as production levels and raw materials,the surface of steel often has defects,which will seriously affect the quality of steel.With the advent of the era of information intelligence,the "Made in China 2025" released by the State Council in2015 pointed out that artificial intelligence is a major strategy to enhance national competitiveness and safeguard national interests and security.Therefore,traditional iron and steel enterprises must use artificial intelligence technology to complete industrial upgrading as soon as possible and improve product competitiveness.In this context,this paper conducts an in-depth study on the defect detection algorithm based on deep learning,and makes targeted improvements to the latest target detection algorithm YOLOv5 according to the characteristics of steel surface defects,which improves the detection speed and accuracy.The main work is as follows:(1)Aiming at the problems of low detection accuracy and slow detection speed in traditional methods for detecting surface defects of steel plates,an improved YOLOv5 s algorithm was proposed.First,use the K-means algorithm based on IOU metric distance to re-cluster the steel data set to obtain multiple sets of anchor boxes;second,Integrate Mix Up on Mosaic data enhancement to suppress over-fitting and improve the generalization ability of the model;then,the network structure is improved and the attention module is integrated to further improve the feature extraction ability of the network;finally,in order to improve the model’s attention to difficult samples The Focal loss is integrated into the loss function to improve the convergence speed and detection accuracy of the network.The experimental results show that m AP of the improved YOLOv5 s algorithm on the test set can reach 78.4%,which is 3% higher than the base network YOLOv5 s.(2)In order to improve the detection speed of steel surface defects,a light-weight channel attention module is proposed,which can effectively focus on important channels with only a small computational cost;secondly,a light-weight spatial attention module is proposed to expand the receptive field by using hollow convolution module,which can extract valuable information in the spatial dimension.For the multi-scale target problem,the multi-level pooling is used to scale the feature maps,and the spatial attention module is used to learn the spatial interdependence information on the feature maps of different resolutions,and a pyramid multi-level attention structure is proposed.Combining a light-weight attention module and a pyramidal multi-level attention structure,a YOLO-Steel steel surface defect detection network is proposed.The experimental results show that compared with the YOLOv5 s algorithm,YOLO-Steel does not increase the time complexity,only a very small increase in the space complexity,and the detection accuracy is effectively improved by 1.8%.