A Road Waterlogging Segmentation Method Based on Improved YOLOv8n-Seg

With the increasing frequency of extreme weather events, urban road waterlogging has become a serious issue, significantly affecting traffic safety and daily commuting. To address the challenges of high computational complexity and insufficient detail feature extraction in road waterlogging segmentation tasks, this study proposes a lightweight segmentation method for road waterlogging based on an improved YOLOv8nseg, named CL-YOLO. The method replaces the original Bottleneck structure with a newly designed C2f-DWR module, which incorporates dilated convolutions to expand the receptive field and enhance the model's ability to capture detailed features of water accumulation. Additionally, a Lightweight Shared Convolutional Detection head (LSCD) is introduced, which leverages parameter sharing and group normalization to achieve efficient multi-scale feature fusion while reducing computational costs. Experimental results demonstrate that the improved model achieves an accuracy of 93.6% and an mAP50 of 93.1% on the test set, with a 25.0% reduction in the number of parameters (down to 2.4M) and a 16.5% decrease in computational cost (down to 10.1 GFLOPs), significantly outperforming mainstream models such as YOLOv8-seg and YOLOv9-seg. This study provides an efficient and accurate solution for real-time monitoring of road waterlogging and holds significant practical value for enhancing urban flood prevention and safety.