Research And System Implementation Of Metal Surface Defect Detection Method Based On Machine Vision

Due to the influence of various factors in the production process of metal industrial products,there may be some surface defects on the surface of metal workpiece.This reduces material strength,shortens workpiece life and increases safety risks.Therefore,it is necessary to carry out quality inspection on the surface of metal products,which is also the key link to ensure the quality of industrial production.Compared with traditional manual inspection,the surface defect detection method based on machine vision has the advantages of high speed and high precision.In the thesis,the surface defect identification of metal plug is studied.Aiming at the problems of low quality and complexity of metal surface defect image,the effective methods of image preprocessing,defect classification and defect target detection in machine vision are explored.Finally,a high speed and high recognition precision metal surface defect detection system is established.In the aspect of image preprocessing,the processing effects of four kinds of filter based denoising methods are compared on the basis of analyzing the image noise.The results show that the median filtering method has the best processing effect and the fastest denoising speed.Next,the edge detection method is used to segment the ROI of the metal surface image,and the pretreatment operation is completed.In the research of defect classification and recognition method,a defect recognition method based on connected domain analysis is proposed for the preprocessed metal plug surface image.The results show that the overall classification accuracy is 89.60% on the original image data set of metal plugs.Then,the surface defect classification methods based on PSO-SVM model and Ghost Net model are proposed to solve the problem that the connected domain analysis method has a small applicable scope and relies on manual threshold setting.At the same time,the NEU-CLS data set is added to compare the defect classification and recognition ability of different methods.The results show that the method based on Ghost Net model has a good ability of defect classification,and the classification accuracy is up to 99.43%.In the research of defect target detection method,NEU-DET data set is introduced.Aiming at the problems of YOLOX-S model in defect data set detection,the optimization measures of improving loss function and adding attention mechanism module are proposed.The results show that the YOLOX-S model using EIOU?Loss loss function has the best target detection performance index on NEU-DET data set.The m AP of the model reaches 79.17% and the detection speed is 71 fps.The surface defect classification method and target detection method are applied to the self-made metal plug data set.The classification accuracy reaches 99.44% and m AP reaches 93.34%.Finally,a detection system is established with image acquisition equipment to meet the requirements of metal plug surface defect detection at multiple levels.The detection system can identify the surface defects effectively.