Structure-Guided Mixed Masked Pretraining and Spatial Continuity Regularization for Printed Circuit Board Defect Detection
Researchers have developed a novel two-phase framework for detecting defects on printed circuit boards (PCBs). The method utilizes structure-guided masked pretraining on unlabeled PCB images to learn structural priors, followed by fine-tuning with spatial continuity regularization for improved defect localization. This approach demonstrated strong performance on the DsPCBSD+ dataset, achieving 85.5% mAP0.5 and 52.3% mAP0.5:0.95, outperforming existing baseline detectors. AI
IMPACT This method could improve automated optical inspection systems for electronics manufacturing.