A Small Target Detection Algorithm for Safety Protection Devices in High-Altitude Work

To address the limitations of existing object detection algorithms in complex backgrounds, small target recognition, and occluded target detection, this paper proposes an improved BPS-YOLO model. First, a Spatial Pyramid Multi-scale Common Convolution (SPMCC) module is designed to enhance feature extraction capabilities and improve detection accuracy for occluded targets. Second, a BiFormer attention mechanism is introduced in the Neck component, leveraging dynamic sparse attention to optimize small target detection performance. Finally, a Soft-CIoU loss function is proposed to enhance bounding box regression for small targets and improve model localization accuracy. Experimental results demonstrate that BPS-YOLO achieves superior performance on a self-collected high-altitude electrical grid operation dataset compared to mainstream object detection algorithms, with anAP m50 of 87.1%, representing a 1.7% improvement over YOLOv11s. This study shows that BPS-YOLO is well-suited for safety protection detection tasks in high-altitude electrical grid operations and provides strong support for safety monitoring in related fields.