Helmet-Wearing Detection Algorithm Based on Improved YOLOv5s

In high-risk industries such as safety production and aerial work, the use of helmets by workers constitutes a crucial measure for preventing injuries. Automated detection of helmet-wearing can enhance objectivity and reduce labor costs. However, the current object detection algorithms for helmet-wearing may overlook small-sized targets, leading to missed detections. To address this issue, this paper proposes an improved algorithm for helmet-wearing detection based on YOLOv5s. First, the attention mechanism is introduced into the multi-scale fusion module of YOLOv5s to enhance the network's expression ability for small target features. Additionally, the Path Aggregation Network (PANet) is replaced by the Weighted Bidirectional Feature Pyramid Network (BiFPN) to improve the network's computational capability and fuse more feature information. Finally, the GIoU Loss function in the algorithm is replaced by the CIoU Loss to make the network converge faster and obtain higher accuracy in regression localization during training. The experiment results show that the mAP and FPS of the improved algorithm have increased by 2.6% and 2.7 frame/s. The improved algorithm can meet the requirements of accuracy and real-time performance for helmet-wearing detection in practical application environments.