Research On Detection And Classification Of Steel Plate Surface Defects Based On Machine Vision

Steel plates are widely used in our lives,and are widely used in some automobile factories,light industrial manufacturing,railway transportation and other engineering fields,which means that the supervision of steel plate quality becomes very important.For small steel enterprises,their detection methods for steel plate surface defects are still relatively backward,and these methods will appear in many problems,not only is it difficult to extract the characteristics of steel plate surface defects,but also will receive interference from external factors,resulting in wrong inspection and missing inspection.To solve these problems,this thesis studies the surface defect processing technology of steel plate based on machine vision.The main research contents and achievements are as follows:(1)This thesis introduces the GoogLeNet network and improves it.Firstly,the spatial attention mechanism and channel attention mechanism CBMA are added to the network to improve the recognition accuracy of the network for steel plate surface defects;Then the Re Lu activation function in the network is replaced by the Leaky Re Lu activation function to solve the problem of neuron death in the process of network training.The improved GoogLeNet network has a good recognition effect in the experiment,and the average accuracy of steel plate surface defect recognition has reached 96.8%,which is 10.2% higher than the initial network model.(2)This thesis continues to study the effect of target detection network on steel plate surface defect detection,and can accurately locate the location of defects,so as to introduce YOLOV5 network and improve it.Firstly,Ghost module is added to the network to extract the feature information of steel plate surface defects;Secondly,CA attention mechanism is added to the backbone of the network to improve the recognition accuracy of the network for steel plate surface defects;Finally,feature pyramid network Bi FPN is introduced,and the ADD operation is added to YOLOV5 network to enhance the feature fusion of steel plate surface defects.The experimental results show that the average accuracy of the improved network for steel plate surface defect recognition reaches 98%,which is about 8% higher than the original network.In this thesis,according to the current situation of steel plate surface defect detection and recognition,two detection methods are proposed,which have high recognition accuracy,will not be interfered by external factors,and the detection speed is fast,and have a very good application prospect in the steel plate production process.