FSAB-YOLO: A YOLO model with fixed-size anchor boxes for high-accuracy positioning of PCB electronic components

The assembly quality of electronic components on printed circuit boards (PCBs) is critical to the performance of electronic products, requiring highly precise component positioning for effective quality analysis. However, conventional object detection methods often produce prediction boxes with fluctuating dimensions, which introduces localization errors and hinders subsequent inspection. In this paper, we propose FSAB-YOLO, a novel detection model specifically designed for high-precision PCB component positioning. To address the instability of box scaling, we introduce a fixed-size anchor box mechanism that incorporates component dimensions as prior information and restricts adjustments solely to the center point. This ensures consistent output dimensions aligned with physical specifications. Furthermore, to capture the structured layout of PCBs, we integrate a criss-cross attention mechanism to adaptively aggregate contextual information along both horizontal and vertical orientations. Complementing these architectural changes, we propose an innovative label assignment strategy to enhance training stability. Experimental results on a real-world PCB dataset demonstrate that FSAB-YOLO significantly outperforms state-of-the-art detectors. Specifically, our model achieves a 3.51% improvement in average precision compared to YOLOv11, highlighting its superior capability for precise localization in industrial automation.