I. Introduction
Automated optical inspection (AOI) with image processing has been studied with the intent to detect the soldering defects on a printed circuit board (PCB) [7]. Much research was related to AOI with traditional ruled-based, image processing, or shallow learning methods for PCB defect detection [8]–[10]. However, most of them did not provide sufficient detection rates for practical applications. Since 2012 when the AlexNet has been developed [11], most AOI designs using convolutional neural network (CNN) methods for image classification and detection application are facing the challenge of how to increase the detection rate and decrease the FAR [12]–[15], [27], [28]. The following shows some novel studies with AOI on PCB defect detection.