To address the issues of limited datasets, tiny defect scales, and complex backgrounds in defect detection for CPCB in industrial settings, a CPCB defect detection method based on PXD-YOLO11s is proposed. Firstly, a high-quality dedicated dataset for CPCB defects is constructed, with a systematically designed acquisition scheme covering 14 types of typical defects such as deep scratches, sand holes, and open circuits. High-resolution defect images are captured using industrial cameras and professional optical equipment, and image preprocessing and data augmentation strategies are employed to enhance sample diversity and generalization. Secondly, improvements are made based on the YOLO11s network architecture: A parallel feature extraction module (ParNet) is introduced, which captures multi-scale feature information through a multi-branch convolutional structure while optimizing the convolution configuration to improve feature extraction efficiency; A dedicated small target detection layer (XsHead) is added to strengthen the recognition capability for tiny defects; the Soft-NMS mechanism is integrated in the post-processing stage to handle overlapping prediction boxes through confidence decay instead of direct suppression, effectively enhancing the detection performance for densely arranged defects; Finally, the loss function is adjusted by adopting DIoU loss to replace the traditional CIoU loss, enabling the model to focus more on the center distance and aspect ratio between prediction boxes and ground truth boxes, thereby improving target localization accuracy in complex backgrounds. Experimental results show that on the self-built CPCB dataset, the improved PXD-YOLO11s model achieves 5.2% and 8.5% increases in mAP50 and mAP(50~95) compared with the original YOLO11s, respectively. In addition, this method outperforms typical algorithms such as Faster R-CNN and YOLOv5s on the public PKU-Market-PCB dataset, demonstrating its excellent generalization ability and effective feature extraction performance. The proposed method significantly improves the CPCB defect detection accuracy and provides an efficient and reliable solution for intelligent industrial defect detection.