AMI-YOLO: a personal protective equipment detection method for multi-miner

Mine safety has always been a key regulatory focus for the government. Research shows that over 90% of coal mine production safety accidents are directly related to human unsafe behaviors, and wearing personal protective equipment (PPE) can effectively mitigate safety risks. This paper proposes AMI-YOLO, a multi-miner full PPE detection algorithm based on YOLOv8n. First, an adaptive weighted fusion module is introduced to achieve dynamic weighting in feature fusion, enhancing adaptability to segmentation and detection tasks in low-light conditions. Next, a multi-fusion multi-scale feature network is designed, integrating multi-path convolutional feature fusion and cross-stage feature lightweight design to enable the model to extract more complete feature information even in complex backgrounds. Third, the CIoU loss function is replaced with Inner-WIoU, further improving the model's detection accuracy for occluded objects and small targets. The proposed algorithm was validated on two self-collected datasets, demonstrating its effectiveness. Experimental results show that under occlusion scenarios, AMI-YOLO achieves improvements of 4.5% and 7.4% in mAP50 compared to YOLOv8n and YOLOv12, respectively. The method meets practical detection requirements and promotes innovation in intelligent mining. To facilitate reproducibility and further research, the source code of AMI-YOLO has been released at: https://github.com/970334745/AMI-YOLO.