Research on the Detection Method of Safety Helmet by Introducing Attention Mechanism

In this paper, YOLOv5 is optimized to maintain high performance in helmet detection. This algorithm incorporates the coordinate attention mechanism to enhance the attention of the network to the identification of small objects such as hard hats. The sensitivity of IoU to small deviation of target position is solved by using normalized Wasserstein distance loss function. Furthermore, the original data is expanded to improve its generalization capability and compel the model to acquire more robust features. The proposed method is specifically applied to helmet-wearing detection scenarios. Test results demonstrate that compared with traditional YOLOv5, the improved algorithm exhibits an increase of 4.1 percentage points in accuracy and 3 percentage points in mAP values for helmet detection. Additionally, there is also a noticeable improvement in recall rate, indicating its effectiveness in detecting helmets across various scenes and scales.