Research On Feature Extraction Of Metal Plate Defects Based On Image Information

Metal plate is the basic raw materials of transportation, industrial production, containers, etc.And, in the country’s economic development, it is very important. With he increase of years in service of metal plate, it will generate a lot of corrosion defects in the natural environment. In addition, coupled with the aging of the metal plates and other reasons, the metal plate will face an accident-prone period and its security reduces quickly. Therefore, the metal plate NDT, to detect defective condition of the metal plate, to make the evaluation about its halthy and to propose maintenance recommendations, it is very necessary which can bring huge economic and environmental benefits. So the research on feature extraction of metal plate defects based on image information is extremely important.MFL picture of metal plate during data acquisition due to electromagnetic interference and other causes will generate image noise, edge blur and low contrast problems. In order to obtain accurate information of defect characteristics, the paper puts forward the corresponding image preprocessing methods. First, through the image denoising and image enhancement methods to improve image quality of MFL picture of metal plate. Second, through image binarization threshold defect extraction method to extract the region of metal plate defect. At last, through the experimental simulation, the pretreatments method to metal plate image is proved effective.Edge detection of the image is the key step in feature extraction of metal plate defects. The thesis use traditional edge detection algorithm for the simulation and analysis, but the traditional method is not very satisfactory. Therefore, the thesis propose PCA-based edge detection algorithm. On the basis of image preprocessing, using PCA algorithm to extract the edge of defect and regions of defects. Since the edge of defects is so wider that coordinates is not clear, the edge of defects need refinement. Finally, we use the eight-connected chain code method to label defects, and gain coordinate information of edges. Through the simulation, the purposes of the boundary determination and boundary information extraction are achieved.In order to identify and describe the geometry of defect information, some defect geometry parameters is defined in the thesis. In addition, deriving the formula of geometric parameters. According to the edge of the chain code defects and coordinate information, calculate numerical information of five defects characteristic parameters of each metal plate defect including the area, circumference, long shaft length and short shaft length, dispersion degree and eccentricity degree. The purposes of detection feature extraction are accomplished. And the result is satisfying. Finally, based on the work of the thesis,the future direction of research and work are prospected.