Resumen
The contribution of this paper is to propose a dual-stream convolutional neural network (CNN) using two solder regions for inspections of surface mount technology (SMT) assembly defects. We extract two solder regions from a printed circuit board (PCB) image and input them to a dual-stream CNN that is constructed for defect classification. The proposed method helps manufacturers efficiently classify and manage defects in an automatic optical inspection system in the SMT line. In addition, since the proposed method uses the solder region for inspection, it can be applied to the inspection of all components having a solder region mounted on the PCB.