UF-Net: Unified Feature Network for Remote Sensing Image Dehazing

Remote sensing image dehazing enhances visibility and supports reliable Earth observation. To overcome the limited global representation of CNNs, the weak sensitivity of state space models to local details, and the underutilization of frequency-domain cues, we propose a unified feature module that integrates global, local, and frequency representations. Based on this module, we design UF-Net, which combines the global modeling capacity of state space models with the local feature extraction of CNNs, while jointly exploiting spatial and frequency domains. This design enables effective global-to-local representation learning and accurate detail restoration. Extensive experiments confirm the effectiveness of UF-Net and its competitive performance against existing methods.