Detection Method of Typical Defects in Arc Ferrite Magnet Surface

An automatic detection approach was proposed to solve unstable accuracy problem of bare-eye inspection of surface defects on arc magnets. According to the geometry features such as the length and area of arc magnet contours, a primary classification of defects was implemented by the support vector machine (SVM), using contour matching similarity as the feature vector. Then, the minimum mean square error classifier was used for secondary classification based on the number and area of detects acquired from analysis of convex and concave defects. The final decision was made by performing the AND operation on the two classification results. The experiments show that the proposed method can achieve an overall accuracy rate of about 91.80%, a fault acceptance rate of about 0.75%, and a correct rejection rate of about 14.00%.