基于YOLOv5改进的铁路工人安全帽检测算法研究

目前铁路上普遍采用人工监督方式来检测工人是否佩戴安全帽，但监督范围过大，在实践中不能及时跟踪和管理所有工作人员。因此针对该问题，采用深度学习目标检测的方法，通过改进YOLOv5s目标检测算法来实现铁路工人是否佩戴安全帽和穿戴背心。具体来说，以YOLOv5s算法为基础，采用GhostNet模块替换原始网络中的卷积Conv，提高模型的实时检测速度；采用更高效简单的多尺度特征融合BiFPN，使特征融合方式更加简单高效，以提高检测速度和降低模型复杂度；把原始的CIOU损失函数替换为SIOU损失函数，以提高模型精度。研究结果表明，改进的YOLOv5s-GBS算法的准确率和识别效率可达到95.7%和每秒45帧，并且模型大小减少了一半，准确率提高了4.5%。

Research on the detection algorithm of railway worker"s hard hat based on YOLOv5 improvement

At present, manual supervision is generally used on railways to detect whether workers wear safety helmets, but the scope of supervision is too large, and in practice it is impossible to track and manage all workers in a timely manner. Therefore, in response to this problem, the method of deep learning target detection is adopted, and the YOLOv5s target detection algorithm is improved to realize whether railway workers wear hard hats and vests. Specifically, based on the YOLOv5s algorithm, the GhostNet module is used to replace the convolution Conv in the original network to improve the real-time detection speed of the model; the more efficient and simple multi-scale feature fusion BiFPN is used to make the feature fusion method simpler and more efficient to improve Detection speed and reduce model complexity; replace the original CIOU loss function with SIOU loss function to improve model accuracy. The research results show that the accuracy and recognition efficiency of the improved YOLOv5s-GBS algorithm can reach 95.7% and 45 frames per second, and the model size is reduced by half, and the accuracy rate is increased by 4.5%.