Research On Real-time Band Steel Surface Defects Inspecting System Based On Machine Vision

With the rapid development of technology, surface quality of band steel is more and more important, which is an important index of the band steel quality. Under the intensely competitive conditions nowadays, it not only stands for the reputation of corporate, but also is the most important elements to hold an advantageous market position. Therefore, the technology of detecting band steel defect effectively and efficiently has attracted much attention of the researchers. An intelligent automation age is coming as soon as the machine vision technology is applied to the band steel defect detection.The thesis adopts Gaussian filter technique to smooth band steel image so that the image is enhanced by means of reducing the noise. Canny algorithm with self-adaptive threshold is adopted to extract the image edge, which is improved on the basis of traditional Canny algorithm to avoid setting the threshold. Due to the use of morphology in image processing, the defect edge will be clearly outlined in the procedure of edge detecting. Then, contours finding technique is made use of to find the contours for the binary image processed above, and partition the contour with biggest area from the image. Finally, BP neural network method with some defect features as input is adopted to recognize the band steel defect.The experimental result demonstrates that the research can detect and recongnize the surface defect of band steel effectively and efficiently. The accuracy and real-time have attained the results expected.