Robust Real-Time PCB Defect Detection Using Improved DETR

This paper presents a PCB defect detection system utilizing the Real-Time Detection Transformer (RT-DETR) model. Originally designed for general object detection, we adapt RT-DETR for the specific task of PCB defect detection, achieving both high accuracy and rapid inference. Building on this baseline, we introduce re-parameterization techniques to further enhance the model's performance with less parameters computational cost. Additionally, we employ data augmentation strategies to enrich the diversity of the dataset, thereby improving the model's robustness and generalization capability. Through these efforts, the developed PCB defect detection system based on the RT-DETR model meets the expected requirements in terms of high accuracy and real-time inference speed. The code and models will be made publicly available.