Abstract: To accurately identify the safety hazards at hydropower construction sites in real time, this paper combines the channel attention mechanism and spatial attention mechanism, improves and applies the YOLOv8 algorithm, and develops an automatic identification method of safety hazards in hydropower construction based on the dual attention mechanism. First, based on the YOLOv8 network framework, we construct a channel attention mechanism to highlight key features adaptively, strengthen dynamically the expression of image features of hidden danger areas, and suppress the influence of background noise. Then, a spatial attention mechanism is built that helps weight important regions, reduce background interference, and optimize feature fusion. It allows to adjust attention adaptively, enhance local detail capture and the positioning accuracy, improve the multi-scale target detection ability, and enhance the spatial feature representation ability of the model. Finally, we verify the accuracy and reliability of the model through a case study of an ongoing construction project. The results show that the proposed method identifies the hazards effectively against the interference in the construction site through the attention mechanism, and achieves an accuracy rate of up to 86.2%, better than previous identification models, thereby improving the dynamic management, prevention and control of hydropower construction safety hazards.

Key words: hydropower engineering, construction safety, hazards identification, attention mechanism, deep learning