Application Research of Improved YOLOv5 Model Based on Multi-Scale Spatial Convolution for Safety Helmet Recognition

In construction sites, safety helmets are critical protective equipment that significantly contribute to workers' head safety. Compliance with safety helmet regulations is an indispensable part of on-site safety management. Existing safety helmet detection algorithms suffer from limitations such as insufficient detection accuracy and poor robustness in complex scenarios. To address these issues, we propose an improved YOLOv5 algorithm for detecting workers and their safety helmets, enhancing detection accuracy. Specifically, we modified the CBL module in the YOLOv5s model by introducing a multi-scale convolutional network to extract spatial features of varying granularities. Additionally, we incorporated a channel-spatial attention module into the Neck section to enhance discriminative feature extraction. Furthermore, we replaced the original loss function with SIoU to optimize detection accuracy and robustness. These IoU variants better address class imbalance and irregular object shapes, thereby improving target detection performance. The proposed YOLOv5-Pro model achieves a mean Average Precision (mAP) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 5. 4 \%}$</tex> on the validation set, surpassing the original model by 6.6%. Experimental results demonstrate that the enhanced YOLOv5-Pro exhibits robust performance under both daytime and nighttime conditions.