PCB Defect Detection Algorithm Based On YT-YOLO

The quality of printed circuit board (PCB) has an important impact on electronic products. Aiming at the defects of PCB, this paper proposes a lightweight detection algorithm model YT-YOLO. Part of the dataset consists of PCB defect data publicly released by Peking University laboratory. SRGAN and data augmentation are used to increase the sample feature granularity and eliminate background noise, respectively. The designed YT Block is used to replace the original architecture to strengthen the feature Extraction ability. Compared with the original model, the parameters are reduced by 16.5%, the prediction accuracy is achieved by 93.5%, and the detection speed is improved by 13.4%. It can be directly deployed in the application terminal with limited computational power. It makes it possible to replace manual quality inspection with automatic, efficient and accurate inspection in the whole process.