Defect detection in printed circuit board assemblies (PCBA) is essential for ensuring product reliability and improving production efficiency in electronic manufacturing. This study proposes GSS-YOLO, an improved YOLOv5-based model that is suitable for defect detection in PCBA. We introduce GSConv of standard convolution and implement VoV-GSCSP in C3 block of the neck to improve feature extraction, enabling better capture of fine details while reducing computational complexity. Additionally, we improve the prediction head by integrating the shuffle attention (SA) module to enable multichannel feature fusion and improve the prediction. These enhancements resulted in a mean average precision (mAP) of 94.0% and an inference speed of 160.24 frames per second (FPS) with fewer than 1.63 million parameters, balancing efficiency, and performance for real-time defect detection. The proposed improvements demonstrate significant improvements in accuracy and computational efficiency, contributing to more reliable and automated PCBA inspection systems.