Abstract: To reduce the incidence of worker safety accidents and effectively enhance safety awareness, this paper proposes a helmet detection algorithm for coal mine scenes, named HFD-YOLO. Using the YOLOv11n object detection algorithm as the baseline, the following improvements are introduced. First, a C3k2-PIPA structure is designed to enhance the model's feature extraction capability for helmets, addressing the challenges of small helmet targets and easily disturbed features. Second, a CCCM-Concat structure is designed to fuse deep and shallow feature information, enabling efficient feature interaction between deep and shallow layers. Local features are used to complement global information, thereby avoiding information loss caused by network deepening. Finally, an MCIM-P2Detect structure is designed to improve the model's multi-scale feature detection capability, enabling it to better adapt to object detection under varying scale transformations. Experimental results demonstrate that the improved model achieves a 4.8% increase in mAP@0.5 compared to the baseline model, reaching 80.2%, which meets the practical application requirements for worker helmet detection.