Integrating multi-scale decoupled representations for traffic image restoration under multiple adverse weather conditions

In real-world environments, data collected by traffic monitoring equipment is often affected by adverse weather conditions, which pose challenges for subsequent intelligent transportation tasks. Although there is currently a wide range of methods available to address image restoration issues in adverse weather conditions, these methods often lack targeted strategies when dealing with composite weather degradation, and it is difficult to balance image restoration and vehicle feature preservation. To address this challenge, this study proposes an image restoration model for traffic scenarios under adverse weather conditions, called the Decoupled Representation Transformer (DRT). DRT achieves precise separation of vehicle features from weather interference through a multi-feature collaborative decoupling mechanism, which consists of a multi-scale Transformer framework and hierarchical decoupled representation modules. Experimental results on multiple synthetic and real-world benchmarks demonstrate that the proposed method achieves excellent performance, with significant improvements in PSNR and SSIM indices. Further experiments in anomaly detection demonstrate that images processed by DRT can improve the accuracy of anomaly event identification, meaning that DRT effectively improves the reliability of image-based traffic information processing and analysis.