Road Object Detection in Foggy Complex Scenes Based on Improved YOLOv10

– Road object detection in adverse weather conditions remains a critical challenge for autonomous vehicle systems and intelligent transportation networks. Foggy environments significantly degrade visual perception capabilities, leading to reduced detection accuracy and increased safety risks. This research proposes an enhanced YOLOv10 architecture specifically designed to address object detection limitations in foggy complex road scenarios. The proposed methodology integrates three key improvements to the standard YOLOv10 framework. First, we introduce a multi-scale attention mechanism that adaptively weights feature maps based on fog density estimation, enabling the network to focus on relevant visual information while suppressing fog-induced noise. Second, a specialized preprocessing module incorporating atmospheric scattering model inversion is implemented to enhance image contrast and visibility before feature extraction. Third, we propose a modified loss function that incorporates uncertainty quantification, allowing the model to better handle ambiguous detections common in low-visibility conditions. Our approach addresses the fundamental challenges of fog interference through a comprehensive analysis of atmospheric degradation effects on object appearance and visibility. The enhanced architecture maintains computational efficiency while significantly improving detection performance across varying fog densities. The methodology combines traditional computer vision principles with deep learning advances, creating a robust solution for real-world deployment scenarios. Experimental validation demonstrates substantial improvements in detection accuracy compared to baseline YOLOv10 implementations. The proposed system shows enhanced capability in identifying vehicles, pedestrians, and traffic infrastructure under simulated foggy conditions. Performance metrics indicate improved precision.