DALENet: degradation-aware lightweight edge-enhanced network for remote sensing image super-resolution

In optical remote sensing and photogrammetric applications, spatial resolution degradation caused by coupled off-axis optical aberrations, platform attitude and orbit perturbations, and atmospheric effects severely undermines imaging quality and geometric measurement accuracy. Existing remote sensing super-resolution (SR) methods rely predominantly on bicubic-downsampled low-resolution (LR) data, which fails to capture realistic degradation and limits practical deployment, while inadequately exploiting high-frequency details and joint global–local contextual modelling, resulting in over-smoothed reconstructions with blurred edges. To address these limitations, this article proposes a Degradation-Aware Lightweight Edge-enhanced Network (DALENet). A Physical Degradation Synthesis Model (PDSM) first estimates blur kernels from real high-resolution (HR) remote sensing images, synthesizing training pairs that closely approximate actual imaging conditions. A Lightweight Edge-enhanced Generative Adversarial Network (LEGAN) built upon Residual Nested Dual-Attention Groups (RNDAGs) captures multi-scale spatial–channel correlations through local and global dual-attention mechanisms, while an Explicit Gradient Prior Block (EGPB) explicitly regularizes the image gradient field as a parallel branch to preserve edge sharpness. An edge-aware loss further constrains edge gradient magnitudes and multi-scale structural similarity, enhancing the reconstruction fidelity of high-frequency components. Experiments demonstrate that DALENet achieves a superior accuracy–complexity trade-off over state-of-the-art approaches.