Abstract: Aiming at the printed circuit board (PCB) defect detection process, which is prone to leakage and misdetection due to the inclusion of abundant small-target defects, an improved EP-RTDETR-based small-target PCB surface defect detection algorithm is proposed. Firstly, CSPDarknet is used to replace the original backbone network to enhance the feature extraction capability of the model. Secondly, the SMPCGLU module is designed to modify the Bottleneck in C2f, which enhances the gated modulation capability and spatial adaptability of the features. Thirdly, the learnable position coding is introduced to improve the scale-interaction mechanism and to enhance the responsiveness to different positional information. Fourthly, small target enhanced pyramid structure small object enhance pyramid (SOEP) is designed based on cross-channel feature mixing (CCFM) module, and the enhanced feature layer and fine feature fusion enable the model to locate and identify the small targets more accurately. Finally, MPDIoU (minimum point distance-based intersection over union)+NWD (normalized wasserstein distance) loss is designed, which accelerates the convergence speed of the model while focusing more on the small target defects, and the regression results are more accurate. The experimental results show that compared with the benchmark model, P improves by 4.6%, R improves by 5.1%, mAP₅₀ improves by 4.6%, the amount of parameters reduces by 16.38 m, the number of floating points reduces by 48.3, and the FPS improves by 8.51, which is able to perform better small target PCB surface defect detection.