Image Detection On Surface Crack Of Billet

In the process of the billet production, various defects affect the quality of the billet and bring great economic losses, so the billet surface crack detection technology becomes one of the keys to improve the productivity of steel and enterprise competitiveness. Traditional detection methods are not applied to the high temperature, so the detection technology based on machine vision is the development trend in the future. Machine vision detection technology is relatively mature in foreign countries, and the detection system has been used in the production line of billet, but domestic research in this is still in infancy, as result, the market competitiveness of China’s iron and steel products has been affected, so the technology is in urgent need of research.The theory of digital image processing on hot billet crack detection based on machine vision is expounded and the hardware structure of detection system is introduced in the thesis. The specific steps of crack detection on hot billet are as follows: first, images of billet are obtained by CCD camera, as the billet keeps beating during production, the locking region method is used to find interested place; then according to the spectral characteristics of the hot billet, threshold segmentation are made based on histogram of interested region and iron filings are inflated; the last step is defection, Hough line detection is used to judge whether there are wounds and Sobel edge detection is used to judge whether there are cracks. Hardware includes the frame structure of the detection system, water cooling of system, separated dust cover, CCD camera and its fixed bracket.The detection method is simple, stable and accurate, it has been verified in the billet production line of Guiyang, as result, the method meets the requirement and has higher application value. But at present, the species of billet surface defects are limited, in order to ensure accuracy of detection, we will be focused on various kinds of defects recognition and classification in the next step.