Objective To address the issues of small targets and insufficient detection accuracy in surface defect detection of printed circuit boards PCBs an improved YOLOv4-tiny model is proposed for PCB defect detection. This algorithm significantly improves detection accuracy while maintaining real-time efficiency. Methods First the second residual block of the backbone feature extraction network outputted a 52×52×128 feature map. Based on this feature map a prediction head named YOLO Head-P3 and a feature pyramid network FPN layer were added. These improvements enhanced the multi-scale feature fusion capability of the FPN layer and solved the structural deficiency of the original YOLOv4-tiny model in detecting small targets. Second an improved spatial pyramid pooling SPP structure was integrated into the FPN to strengthen multi-scale feature fusion thereby improving detection accuracy. Finally the efficient channel attention network ECAnet was introduced. By adaptively adjusting feature weights through an attention mechanism this network further enhanced the adaptive feature extraction capability of the model while reducing complexity and computational demands to meet the precise requirements for small-target defect information. Results Experimental results showed that the improved algorithm achieved a mean average precision mAP of 97. 32% which was 7. 69% higher than that of the original model with a detection speed of 113. 67 frames per second fps . Conclusion Compared with the original model all evaluation metrics of the improved model have been enhanced. The improved model enables real-time monitoring of small-target defects and exhibits excellent performance.