Tunnel Personnel Safety Helmet Wearing Detection Algorithm Based on Improved YOLOv5s

With the rapid development of railway construction, the scale of tunnel engineering continues to expand, occupying an increasingly important position in railway construction. In railway tunnel construction, ensuring that construction workers correctly wear safety helmets is a crucial part of safety management. However, the performance of existing detection algorithms fails to meet the needs of safety management in the complex environment of dim light and heavy dust inside tunnels. Aiming at the construction scenario of railway tunnels, this paper constructs a specialized safety helmet detection dataset and proposes a construction worker safety helmet detection algorithm based on the improved YOLOv5s. By designing a new backbone network structure, ML-CSPDarknet, the network's feature extraction ability in complex environments is significantly enhanced. A lightweight upsampling operator, CARAFE, is introduced to effectively expand the network's receptive field. A coordinate convolution CoordConv module is constructed and integrated into the corresponding positions of FPN and Head, remarkably improving the network's perception ability of spatial information. The results show that compared with the original algorithm, the improved detection algorithm has increased by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2.8 \%, 2.9 \%, 3.5 \%$</tex>, and 2.6 % in precision, recall, and average precision (mAP50, mAP0-95), respectively. Compared with other mainstream detection algorithms, the overall performance of the algorithm in this paper is also more outstanding, which can effectively meet the detection requirements of the wearing status of safety helmets of construction workers in the railway tunnel construction scenario and provides strong technical support for the safety management of railway tunnel construction.