Abstract: As safety helmet provides essential head protection for workers at construction sites, wearing safety helmet is of great significance to their lives. However, workers often do not wear helmets due to a lack of safety awareness. With the continuous development of deep learning technology, YOLO series algorithms with high precision and speed have been applied to various scene detection tasks. In order to establish a digital helmet monitoring system, this paper proposes a method to detect helmet wearing based on YOLOv5. Through the sample expansion method of data enhancement, basic images and data enhancement are used to optimize the data set, and a self-developed data set with rich features about helmet wearing is built. Consequently, the model can accurately identify the wearing condition of helmet and realize the real-time detection. The experimental results show that the average detection speed of YOLOv5 reaches 60 f/s, which can meet the condition of real-time detection. The mAP value reaches 98.5%, which proves the effectiveness of the helmet detection based on YOLOv5.