DAH-TrafficRSNet: Dual-Branch Traffic Remote Sensing Image Dehazing Network Based on Atmospheric Scattering Model and Hierarchical Feature Interaction

Fog degrades the quality of traffic remote sensing images and severely restricts their applications in scenarios such as intelligent traffic monitoring and road network planning and evaluation. This paper proposes a dual-branch traffic remote sensing image dehazing network (DAH-TrafficRSNet), which integrates the advantages of prior-based methods and deep learning to address the clarity degradation caused by fog. The network is based on a dual-branch architecture. One branch incorporates the atmospheric scattering model (ASM), which preliminarily estimates and corrects fog-induced degradation, providing a physical basis for restoring key elements such as road textures and vehicle shapes. The other branch adopts a hierarchical feature interaction (HFI) mechanism, excavating image features at multiple scales and levels to enhance the ability to capture details like airport runway lengths and tarmac partitions in complex foggy environments. Experimental results show that DAH-TrafficRSNet performs excellently on remote sensing image datasets of various traffic scenes. Compared with traditional dehazing methods and some advanced deep learningbased dehazing models, it can remove fog more effectively, accurately restore road connectivity and vehicle integrity, and significantly improve image visual quality as well as the accuracy of analytical tasks such as road condition evaluation. This provides reliable support for the practical application of remote sensing technology in the field of intelligent transportation.