A Safety Helmet Detection Algorithm Based on Improved YOLOv5s

Wearing safety helmets is an important way to ensure the safety of workers' lives. To address the challenges associated with low accuracy, large parameters, and slow detection speed of existing safety helmet detection algorithms, we proposed a safety helmet detection algorithm YOLOv5s-CR based on YOLOv5s. First, we use a lightweight backbone, a high-resolution feature fusion network, and a small object detection layer to improve the detection accuracy of small objects while substantially decreasing the parameters of the model; Then, we embed a coordinate attention mechanism into the feature extraction network to improve the localization accuracy of the object to be detected; Finally, we propose a new receptive field enhancement module (RFEM) to substitute the SPPF module in the original network, enabling the model to acquire features under multiple receptive fields, thereby enhancing the detection precision of multi-scale objects. Using the Safety Helmet Detection dataset for validation, in contrast to the initial YOLOv5s, the parameters of the YOLOv5s-CR were reduced by 62.8%, and P, R, and mAP <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5</inf> were increased by 1.5%, 1.2%, and 2.0%, respectively. The detection speed can reach 149FPS on the RTX3070 GPU, which can satisfy the accuracy and real-time requirements for detecting safety helmets.