In the domain of printed circuit board (PCB) defect detection, challenges such as missed detections and false positives remain prevalent. To address these challenges, we propose a small-sample, high-precision PCB defect detection algorithm, called SSHP-YOLO. The proposed method incorporates an ELAN-C module that merges the convolutional block attention module (CBAM) with the efficient layer aggregation network (ELAN), thereby enhancing the model’s focus on defect features and improving the detection of minute defect details. Furthermore, we introduce the ASPPCSPC structure, which extracts multi-scale features using pyramid pooling combined with dilated convolutions while maintaining the resolution of feature maps. This design improves the detection accuracy and robustness, thereby enhancing the algorithm’s generalization ability. Additionally, we employ the SIoU loss function to optimize the regression between the predicted and ground-truth bounding boxes, thus improving the localization accuracy of minute defects. The experimental results show that SSHP-YOLO achieves a recall rate that is 11.84% higher than traditional YOLOv7, with a mean average precision (mAP) of 97.80%. This leads to a substantial improvement in the detection accuracy, effectively mitigating issues related to missed and false detections in PCB defect detection tasks.

SSHP-YOLO: A High Precision Printed Circuit Board (PCB) Defect Detection Algorithm with a Small Sample