Surface defect detection of printed circuit boards (PCBs) is critical for ensuring the performance and reliability of electronic devices. With increasingly complex PCB designs, defects typically manifest as small target features within complex backgrounds, yet existing PCB defect detection datasets overlook these extremely small feature defects in high-resolution images. To address this issue, we introduce HiPCB, a high-resolution PCB defect detection dataset. Compared to PKU-Market-PCB, the HiPCB dataset contains 2.14 times as many defects per image, while the annotation boxes relative to image size are reduced by more than 10 times, accounting for only 0.071‰. Existing object detection methods perform poorly in these high-resolution small target scenarios, failing to meet practical application requirements. Therefore, we propose YOLO-SKCA, a method based on YOLOv8. This method innovatively integrates sliced inference strategy to solve the feature loss problem in high-resolution images, enhances perception of horizontal and vertical defect features through the Strip-Shape Kernel Cross Attention (SKCA) module, and introduces a Detail Enhancement (DE) module to improve the model’s detection capability for small-sized defects. Experiments show that YOLO-SKCA achieves state-of-the-art performance on HiPCB, with a mAP of 72.1%.