Abstract: Aiming at the importance of defect inspection and the shortcoming of manual inspection in
production process, the automatic machine vision detection and classification method is studied for
the surface defect of arc magnet. Firstly, against many kinds of defects with low contrast, textured
background and uneven brightness, the scan line gradient is defined to constitute feature vector with
the standard deviation of the scan line grayscale. Secondly, the image segmentation method based on
two-class support vector machine is presented to identify and extract defects. Finally the improved
method on multi-class support vector machine is proposed to classify these extracted defects, which
solved the problems of unclassifiable region and improved classification accuracy and effectiveness.
The experimental results indicate that all kinds of defects of the different sub-region can be detected
rapidly and accurately. The detection rate of defects can reach 96% and the classification rate of
defects is higher than 91%.

Key words: machine vision, surface defect, scan line gradient, support vector machine, arc magnet