Abstract
Image processing using neural networks is applied to the inspection of electronic connector plug pins. Some defects among a large number of pins inserted into a board are detected by the proposed system instead of visual observation. Pin images taken by ITV camera are first processed to digital binary signals, then input to the first layer of three-layered neural network. The training of the network is accomplished with the back-propagation algorithm. Each weight interconnecting units of adjoining layers is modified by learning many types of pin conditions. Fundamental features of the proposed system were made clear, and the network could well detect such defects as bended, mis-positioned and buckled pins.
