Abstract
Helmets are the most common protective equipment on construction sites and can effectively reduce head injuries caused by falling objects. In actual helmet detection on construction sites, traditional target detection faces challenges such as complex environments and unclear target identification. To address this issue, we developed the URD-YOLOv8 helmet detection algorithm. The algorithm is based on an improvement of YOLOv8 and aims to enhance the performance of helmet detection. First, the upsampling module is integrated into the neck network of the model, which makes the model upsampling ability improved and make the image more detailed, thus preventing information loss. Second, a novel convolution module is proposed to help the network focus more on important feature information and improve the effectiveness of the model in feature extraction. Finally, propose a new structure of information aggregation, It better fuses information about target characteristics and context at different scales, allowing the information to flow between channels, thus improving the algorithm's performance. Experiments show that the precision, recall, mAP@0.5 and mAP@0.5:0.95 of the improved helmet wear detection algorithm are higher than the original algorithm by 1.07%, 0.58%, 1.18% and 0.95, respectively.