Research On Real-time Inspection System Of Steel-Strip Surface Hole And Relevant Technology

As a newly interdisciplinary technique, image processing and recognition has wide application in industry and manufacture engineering. This dissertation is devoted to the research on image processing and recognition techniques for the detection system of steel-strip surface hole defects. It represents special theories and algorithms according to the specialties of steel-strip images, proposes the solution of hole defect detection and segmentation and subsequent hole feature extraction, selection and recognition, successfully realizes the real-time inspection of steel-strip surface hole defects, and improves the automatic level of producing steel-strip. Analyzing the method of image preprocessing, this dissertation adopts image smoothing method to de-noise in order to enhance the image. Comparing the advantage and disadvantage of the various edge detection operators, this dissertation uses optimal edge operator to extract defect contour. On the base of edge extraction result, this dissertation introduces a fast clustering algorithm for thresholded image segmentation to realize the clustering of hole defect edges, acquire the position of hole defect area and successfully reach the aim of hole defect orientation and segmentation. This dissertation detailedly introduces the principle of feature extraction and feature selection, proposes an improved algorithm of image binary to effectively separate the hole and background region with the interferential feature of intense glisten, analyzes the features of hole defect image of steel-strip surface and discusses the algorithm of feature extraction. Also, this dissertation introduces how to improve the BP network structure and learning algorithm and introduces in detail the process of designing the classifier using improved BP network, based on the detection of steel-strip surface hole defects. Adopting the algorithm of this dissertation to recognize the holes for 500 steel-strip image samples (1024X768 pixels, 256 colors), the system has the recognition precision of above 98%. For one image of 5 holes (30X50 pixels), computation time is between 20ms and 30ms and can satisfy the need of the online real-time detection of steel-strip holes. The system has successfully been applied in producing the steel-strip in Bao Steel in Shanghai. At the same time, it has directive importance on the other fields of image processing and recognition.