Safety Helmet Wearing Detection Method Based on Improved YOLOX

Aiming at the problem that the current safety helmet detection methods are easy to miss and false detect small objects and occluded objects in crowded scene, a lightweight safety helmet detection method based on YOLOX was proposed. Firstly, the coordinate attention mechanism is added into YOLOX backbone network to improve the network's perception of object position. Secondly, a weighted bidirectional feature pyramid was used to replace the feature fusion part of the neck network to achieve simple and fast multi-scale feature fusion, and depthwise separable convolution is used to reduce the number of parameters of the network and model size. Finally, a new detection scale is added to further strengthen the detection ability of small objects. On the helmet wearing detection dataset SHWD, the experimental results show that the mAP of the improved method reaches 92.2%, which is improved by 2.34% compared with the original algorithm, and the model size is reduced by 16%. The missed detection and false detection of small objects and occluded objects are improved.