<p>Deep-learning-based super-resolution (SR) methods have shown remarkable performance, but they cannot reconstruct degraded images with different scale factors and blur kernels with a single model. Deep unfolding networks can be used to handle complex SR tasks with different degradation scenarios using a unified network model by constructing a model-driven neural network. However, existing SR methods based on deep unfolding typically rely on a fixed and manually designed transformation operator to construct prior terms, which limits the ability of feature extraction and the improvement of SR performance. To address these issues, we introduce multiple learnable priors in the construction of deep unfolding network and propose a novel super-resolution reconstruction framework for remote sensing images, termed Multiple Learnable Priors-based Unfolding Network (MLPUN). In the proposed approach, multiple learnable transformation operators are introduced to construct flexible and adaptive prior terms. These priors are integrated into an SR model which is subsequently unfolded into a trainable neural network. This unfolding process enables simultaneous optimization of both transformation operators and prior functions, enhancing the model’s ability to exploit domain-specific information for SR reconstruction. Furthermore, to reduce computational complexity while preserving performance, a lightweight variant named Efficient MLPUNet (EMLPUN) is proposed. EMLPUN significantly reduces the number of network parameters, making it more suitable for deployment in resource-constrained environments without compromising reconstruction quality. Experimental results on three benchmark remote sensing datasets demonstrate the superiority of the proposed methods over existing SR approaches. The proposed models achieve higher accuracy in image reconstruction and exhibit enhanced capability in preserving edge structures and fine details.</p>