A lightweight safety wear detection algorithm for underground workers based on improved YOLOv8

This study presents Star-YOLO, an improved YOLOv8n-based algorithm for detecting underground workers' safety equipment. To address low accuracy and high complexity in mining environments, we introduce three enhancements: 1) Replacing the backbone with the lightweight StarNet network to improve efficiency; 2) Designing a C2f-Star feature fusion module using StarBlock;3) Developing a lightweight shared convolution detection head. In our self-constructed underground worker safety dataset, Star-YOLO achieves a mean Average Precision (mAP@0.5) of 95.0% while reducing parameters by 52.3%, FLOPs by 44.4%, and model size by 54.4%, reaching 384 FPS. Star-YOLO effectively sustains elevated detection accuracy while fulfilling real-time performance and lightweight deployment criteria. Ablation studies validate the efficacy of each enhanced module.