Hybrid pooling and channel attention‐based single‐shot detector for printed circuit board defect detection

Abstract

Automatic defect detection is a key but challenging technology for managing printed circuit board (PCB) production quality. In recent years, deep neural networks (DNNs) have attracted considerable attention for PCB defect detection. However, due to the complexity, diversity, and small-scale characteristics of defects, it remains a challenge. In this letter, a single-shot detector (SSD)-based defects detection method is proposed. This method treats the test and template image as a single image with two channels. In order to reduce the information loss caused by the traditional pooling methods, a new pooling method combining three types of pooling features is proposed. To further improve the detection performance, the channel attention (CA) mechanism is introduced into the detection network. Our proposed method achieves 99.64% mean average precision (mAP) on the DeepPCB dataset, which surpasses the state-of-the-art methods.