MuSAP-GAN: printed circuit board defect detection using multi-level attention-based printed circuit board with generative adversarial network

Abstract

A printed circuit board (PCB) is one of the important components in every single electronic device, which assists in connecting each component for many purposes. Somehow, the PCB can be affected due to spurs, short circuits, mouse bites, and so on. Therefore, the detection strategy for such defects is very important and also complicated. So, this research concentrates on developing a deep learning model, a multi-level attention-based printed circuit board with a generative adversarial network, and a YOLOv5 (MuAP-GAN-YOLOv5) model for defect detection in PCB. The contribution of this research is to enhance image quality using the proposed multi-level attention-based PCB-GAN (MuAP-GAN) method, which is embedded with a multi-level attention mechanism to enhance image quality. Therefore, the model can efficiently learn and train for accurate defection as well as localize the defected area in PCB. Here, the YOLOv5 model plays an important role in training based on enhanced features and, therefore provides accurate results. In addition, this model requires less computational expenses, is quite reliable, also provides a maximum accuracy of 95.24% compared to other traditional methods.