An Adaptive Defect-Aware Attention Network for Accurate PCB-Defect Detection

Defect detection is a critical component of quality control in the manufacturing of printed circuit boards (PCBs). However, accurately detecting PCB defects is challenging because they are very small and inconspicuous. In this article, an adaptive defect-aware attention network (ADANet) is proposed for PCB defect detection, and it contains two main modules: small defect preserving and location (SDPL) and defect segmentation prediction (DSP), where the SDPL module is designed to extract the high-resolution and multiscale defect feature representations to avoid the loss of small defects caused by model depth and then locate their positions with a deformable Transformer, and the DSP module is developed to predict their categories and masks. Experimental results conducted on two PCB datasets show that the proposed ADANet can surpass state-of-the-art approaches and achieve high performance in multiscale defect classification and detection results.