Multi-attention YOLOv5 for Recognition of Abnormal Helmets Wearing in Power Stations

Safety helmets must be worn by people who work in the power plant construction area. In order to automatically monitor whether workers are wearing safety helmets correctly and ensure safe operation in substations, a safety helmet wearing detection algorithm based on improved YOLOv5 named, multi-attention YOLOv5, is proposed in this paper. First, based on the YOLOv5 algorithm, a multi-attention mechanism module is proposed including the spatial attention mechanism and the channel attention mechanism. By redistributing the weights to different features, the semantic features are enhanced by two attention mechanisms, and the recognition ability of the YOLOv5 for small objects is improved. Then a novel loss function is designed, which aims to alleviate the gradient problem and slow convergence problem of the traditional loss function used in YOLOv5. Finally, we propose a dataset collected from various electrical work areas. The results of experiments on the dataset demonstrate the superior potential of our model and it can be used to accurately recognize abnormal behavior of helmets wearing.