Research On Early Detection And Segmentation For Silicon Steel Surface Defects Under Texture Background

Silicon steel is indispensable magnetic alloys for power, electronics and military industry, mainly used in iron core of various motors, generators and transformers. To meet the needs of energy saving and new electromechanical products making, the requirement of the silicon surface quality is higher and higher, while the surface defects influence producing and products passing rate, largely restricted the development of steel enterprise. In the process of manufacturing, steel enterprise is very concerned about how to detect defects in time, by means of the analysis of defects, finding out the causes of the defects is the problem needed to be solved, and the ultimate goal is to complete quality control.In the current technical level, the detection of silicon steel surface is still in the stage of research. As the demand for the surface quality rise, the steel enterprise needs to improve the detection accuracy. Silicon steel surface is full of texture, while the general steel surface is non-textured in the detection process. Therefore, how to detect the silicon steel surface defect is a new research task in the process of producing. The purpose of this study is to realize early detection and segmentation for silicon steel surface defects in texture background.The main achievements of this thesis are as follows:(1) The current detection method based on gray level is not suitable to defect detections of silicon steel sheet under texture background. A method based on spectrum residual is applied to detect the surface detects of silicon steel sheet. The difference of the gray value is the judgment of defects existing.150images are Chose for the experiments; the experimental results show that the detection rate is94%, therefore, the method based on spectrum residual can detect the silicon steel surface defects.(2) The segmentation method based on gray information cannot be used in silicon steel surface segmentation under texture background. In this paper, in order to solve this problem, the active contour segmentation method based on structure tensor is proposed. First of all, the traditional structure tensor was improved, the local information joined the structure tensor, The structure tensor based on region information is proposed, it can extract the texture features better. Then, in the features space extracted by structure tensor, active contour model is set up to segment images based on regional probability density function KL distance; the solution of the model is segmentation results. In order to reduce the computational complexity of this model, a quick Split-Bregman numerical method has been used in solving the model. Finally, the algorithm proposed in this paper is used to segment some of the common detection of silicon steel sheet, such as scratches, eyewinkers, holes. The experimental results show that the proposed algorithm can segment the defection area accurate, and effectively overcome the influence of noise.