Research On Key Technologies Of The Strip Surface Defect Detection

Iron and steel industry is one of the key industries in China.As one of the pillar products of the industry,strip steel has been applied in various fields,such as weapons and equipment manufacturing,automobile manufacturing,large-scale aerospace equipment production and so on.The quality of strip steel products directly determines the performance of products,so it is of great significance to detect the surface defects of strip steel.Under this background,the key technology of strip surface defect detection system is studied in this paper.The image denoising algorithm,image enhancement algorithm,image segmentation algorithm,image feature extraction and selection,defect recognition and the design of defect image database are emphatically studied.The main contents of this paper are as follows:(1)Pretreatment technology for surface defects of strip steel.In order to reduce the noise produced by image acquisition system and acquisition environment,several common filtering and denoising algorithms were used to process defective image.The minimum absolute error and signal-to-noise ratio are used as the evaluation index of denoising effect,and the median filter is selected to denoise the defective image comprehensively.After that,the quality of defect imagebecomes low.Comparing several image enhancement methods,Retinex method is chosen to enhance the image.Several common image segmentation methods are analyzed.Finally,a connected region segmentation method based on gray threshold is selected.By preprocessing the defect images,it can be achieve the purpose of removing the noise of the defect image,enhancing the image quality and segmenting the defect.(2)Feature extraction,selection and classification of strip surface defect image.From the point of view of shape feature,defect details and micro and macro features,the shape feature,singular value feature and texture feature of the segmented image and gray image are respectively extracted.A total of 150 features are extracted from each defect sample.In order to eliminate redundant features,the 75 dimension features with the greatest contribution are selected as the final sample features by using mRMR method.SVM,ElM,Softmax and DBN classification model are selected to validate the classification for the small sample multi-classification problem of steel strip defect image.By analyzing the running time and recognition accuracy of the classifier,SVM isfinally selected,which has the shortest running time and can achieve 98.7% recognition accuracy.(3)Designing the database of strip steel surface defects.Comparing with three common database software,we choose SQL Server 2008 software to design the database of surface defects of strip steel.Because of the powerful digital image processing ability of MATLAB,this paper chooses GUI of MATLAB to design the defect database interface.At the same time,it has completed the design of data entry interface and database table through programming and the interface of surface defect database and achieved the function of fuzzy query and accurate query of image data and preliminarily explores the matching and retrieval of defect image.