Research on Attention Mechanism and Loss Function Based on YOLOv5 to Improve the Detection Method of Underground Personnel Wearing Safety Helmets

Abstract Underground helmet wearing detection has always been a popular field in target detection, due to the dim and complex underground environment, the existing object detection model algorithm is not very ideal for the detection effect of underground helmet wearing, so it is necessary to improve the detection effect of the object detection model algorithm on the wearing of safety helmet in the mine environment. In order to improve the accuracy of helmet wearing detection, the SENet attention mechanism was introduced with the multi-pixel branching structure modification, and the smoother Swish activation function was used to replace the ReLU activation function. Combining the advantages of NWDLoss in small target detection and IoULoss in the case of high overlap between the prediction frame and the real frame, the two were weighted. Experimental results show that the SENet-tn attention mechanism can be inserted into the neck end of the model algorithm to significantly improve the accuracy, and the average detection accuracy can be increased to 89.47%. Compared with the improved model algorithm, the average detection accuracy of the improved model is increased by 2.91%.