Research On Steel Surface Defect Inspection System Based On Computer Vision

The iron and steel has become the indispensable raw materials of automobilemanufacturing, machinery chemicals, ship manufacturing, military national defenseand other industries. However, due to lagging rolling equipments, and theundeveloped process technology, resulting in the formation of roll mark, adhesion,creasing, wrinkling, scratches, oxidation of color and different types of defects in thesteel surfaces, which directly affect the appearance and quality of the steel, so it isimportant to improve the detection level of the steel surface defects.This research project mainly made a thorough research done in-depth researchin the following areas: detection equipment’s non-labor, image processing method,the combination of real-time and time-sharing system, classification and recognition.This is conducive to improve the detection level, ensure the quality of products, inorder to improve the market competitiveness of the final steel products, therebyenhancing the competitiveness of our country’s steel market. The specific researchcontents are as follows:Starting from the steel surface defect detection system, first of all, proposed theprogram of the steel surface defect detection system based on the upper and lowersurface scanning of the linear array CCD, Using image processing algorithmsconducted noise reduction processing and sharpening processing to the collecteddefect image, the purpose is that the useful feature region be separated from thebackground and provided convenience for subsequent feature extraction. Secondly,researched on the knowledge of genetic algorithm, and combined with thecharacteristics of feature vectors, designed feature extraction method using geneticalgorithm. Under the premise of reserving obvious features, optimized the imagefeature set, reduced the dimension of the defect feature, in order to improve theperformance of the machine learning algorithms. At last, Depth study of theapplication of the BP neural network algorithm in the direction of defect detection, and proposed the correlation improved method, designed classifier using of theimproved BP neural network algorithm, and the effectiveness of the method isproved by experiments, and the recognition rate reached90%.