Gradient-Prior-Guided Dual-Branch Network for Preserving Fine Structures in Remote Sensing Image Super-Resolution

In the task of remote sensing image super-resolution reconstruction, both CNN and Transformer have demonstrated impressive capabilities. However, many existing dual-branch models only perform feature fusion through simple operations such as concatenation or weighting, which fail to fully exploit the complementary advantages of CNN and Transformer in feature representation. To address this limitation, this paper proposes a novel Gradient-Prior-Guided Dual-Branch Network (GPG-DBN) for remote sensing image super-resolution. Specifically, the model incorporates an enhanced Transformer structure to capture long-range dependencies and a CNN branch to extract local structural details. To further enhance detail recovery, a Gradient-aware Enhancement Module (GEM) is introduced to guide the network using explicit gradient priors, effectively highlighting high-frequency information such as edges and contours. In addition, we design a Feature Fusion Module (FFM) to deeply integrate the global and local features from both branches, achieving more effective information interaction and complementary learning. Extensive experiments conducted on the UC-Merced and WHU-RS19 datasets demonstrate that the proposed GPG-DBN achieves superior performance in both quantitative metrics and visual quality compared to existing state-of-the-art methods, validating its effectiveness and robustness.