Snow accumulation severely impairs image quality in traffic monitoring and intelligent transportation systems, posing challenges for reliable scene understanding during ad-verse weather. To tackle this issue, we propose the Frequency and Physics-Aware Snow Removal Network (FPA-SnowNet), an innovative architecture that integrates two specialized branches: a physics-aware module that explicitly models snow density, brightness, and motion characteristics, and a frequency-aware module leveraging the discrete wavelet transform to capture fine-grained multi-scale snow textures. This dual-branch design allows FPA-SnowNet to detect and remove diverse snow patterns more accurately while preserving structural details and realistic appearance. Extensive experiments on the Snow Cityscapes dataset demonstrate that FPA-SnowNet achieves state-of-the-art performance, with an average Peak Signal-to-Noise Ratio (PSNR) of 38.82 dB and a Structural Similarity Index Measure (SSIM) of 0.9816 across small, medium, and large snow conditions. In addition, it maintains high computational efficiency with an average inference speed of 62.02 Frames Per Second (FPS). These results highlight that FPA-SnowNet offers a robust and lightweight solution for real-time traffic scene image enhancement in snowy environments, contributing to clearer visual perception and safer transportation operations during adverse weather.