A Lightweight Transformer for PCB Defects Detection

In this paper, a lightweight Printed circuit boards (PCB) defect detection model (LPDETR) is proposed, which maintains high detection performance while effectively addressing the limitations of existing methods, such as large parameter sizes and high computational demands. Motivated by the need for efficient and real-time defect detection in PCB manufacturing, LPDETR builds upon Real time Detection Transformer (RT-DETR) by introducing a streamlined network architecture tailored for resource-constrained environments. Extensive experiments on a public PCB defect dataset demonstrate that LPDETR surpasses various advanced algorithms in balancing detection accuracy and inference speed. The results reveal that, while ensuring high precision, the model size is reduced to 27.3 MB, approximately 30% smaller than the original RT-DETR. This confirms that LPDETR is a competitive and reliable lightweight solution for PCB defect detection, offering practical value for industrial applications where rapid and accurate inspection is critical.