Small sample PCB defect detection based on meta feature enhancement

Aiming at PCB surface defect detection tasks under small sample conditions, a meta learning scheme is introduced to fully extract prior knowledge and quickly generalize on new defects. Firstly, combining meta learning with fine-tuning strategies, only fine-tuning the detector head during the meta testing phase to improve classification ambiguity during knowledge transfer; Secondly, to address the issue of confusion between new and base class defects in PCB, a global feature fusion module is designed in support branches to fuse global channel features with original support features to distinguish different defect categories; Finally, introducing a self attention module on the query branch enhances the network's attention to small targets, helping to solve the problem of missed detection of defective targets. The experimental results show that the proposed method exhibits excellent detection performance in 10 shot tasks, achieving 62.4% mAP in the new class.