RSE-YOLOv11: A Safety Helmet Detection Algorithm for Construction Environments in Rainy and Foggy Weather

Abstract

To address the issues of severe image degradation and high missed detection rates of existing safety helmet detection algorithms under severe weather conditions such as rain and fog on construction sites, an object detection algorithm named RSE-YOLOv11, based on an improved YOLOv11, is proposed. First, a rainy safety helmet adaptive module (RHTDM) is designed to enhance the perception of target contours under low visibility through multi-scale spatial feature extraction and edge enhancement. Second, a cascaded attention mechanism prioritizing spatial information and refining channel weights (SCSA) is introduced to suppress rainy and foggy background noise and reduce feature redundancy. Finally, an upsampling module based on feature reconstruction (EUCB) is employed to recover the loss of high-frequency details caused by rain occlusion. Experimental results on a custom rainy and foggy construction dataset demonstrate that the mAP@0.5 of RSE-YOLOv11 reaches 92.4%, achieving a 5.0% improvement over the baseline, with a parameter count of only 3.34M. While ensuring a lightweight design, the proposed algorithm effectively overcomes missed and false detections under extreme meteorological conditions, meeting the all-weather supervision requirements of smart construction sites.