YOLOv8-Based Multi-Object Detection Method for Adverse Weather Road Conditions

This paper proposes an improved YOLOv8-based object detection model to address low accuracy, missed detection, and false positives in adverse weather conditions. First, a BotNet structure is added to the last layer of the backbone network to enhance feature extraction in challenging weather. Additionally, a Large Separable Kernel Attention (LSKA) module is introduced in the C2f module to reduce parameters while applying visual attention. Finally, CSPNet is replaced with the BIFPN feature fusion network. Experimental results show that the improved model outperforms YOLOv8 by 2% and YOLOv5 by 4.5%, demonstrating its superior performance in adverse weather.