Research on the Identification of Safety Helmet Violations at Tunnel Construction Sites Based on an Improved YOLOV5L Model

Safety is crucial in tunnel construction, and wearing helmets is a basic protective measure for workers. However, the current site relies primarily on manual inspection, which is inefficient and prone to oversight. In particular, accurately and efficiently detecting small targets (e.g., helmets) in complex construction environments remains challenging. To address this issue, this paper proposes a lightweight detection model based on an improved YOLOv5L. This model replaces the backbone network with MobileNetV3, significantly reducing the number of parameters and computational cost. At the same time, it incorporates an attention mechanism and adds a small target detection layer, effectively improving detection accuracy. Experiments based on the self-constructed tunnel construction site helmet dataset, T-Helmet, show that the model's mean average precision (mAP) at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 5}$</tex> is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 3. 2 \%}$</tex>, which is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{2. 6 \%}$</tex> higher than the original model. The detection speed is 31 frames per second. This model significantly improves small target detection performance while meeting the demand for lightweight deployment onsite, providing effective technical support for the intelligent supervision of tunnel construction safety.