At the site of large-scale water conservancy construction projects, there are problems such as falling objects, tower crane rotation, wall collapse, which pose a great threat to the personal safety of construction personnel. Wearing safety helmets is an effective measure to protect construction personnel. Therefore, it is necessary for construction personnel to carry out accurate detection of helmet wearing as a safety management in engineering operations. Aiming at the problems of missing detection and low detection accuracy of small and dense helmet targets in large-scale hydraulic construction scenarios, a helmet wearing detection algorithm based on STA-YOLOv5 is proposed, which introduces Swin Transformer and attention mechanism into YOLOv5 algorithm to improve the identification ability of the model to the helmet. The experimental results show that the STA-YOLOv5 algorithm has more accurate detection results, and the recognition accuracy reaches 91.6%, which is significantly improved compared with the original YOLOv5 algorithm.