Study on the flaw recognition algorithm based on neural network

Based on cognition and research for vision measurement method on chips' defects, a new type of chip defects detection algorithm is proposed. In view of defect characteristic and examination request of chips' surface, the first is pre-processed on the images by CMOS camera. This process includes gradation?median blur?iterative segmentation and contours extraction on images, which will get the chips' contours including targets and backgrounds with stark contrast. On this basis, the chips' online defect detection and classification algorithm including defects extraction and classification is researched. The defects extractions carried on the invariant and geometric positioning features of invariant moments to chips' image correction and then use frame-difference method to extract defects. The defects classification is that a hybrid algorithm based on RBF Neural Network algorithm is applied to achieve the online detection for chips by real-time, fault-tolerant characteristics of neural networks. Above all, it satisfies the requirements of online detection. © 2013 Asian Network for Scientific Information.