基于Mamba架构的遥感图像超分辨率重建方法综述OA

The visual architecture Mamba,built on selective state-space models,enables long-range de-pendencies to be modeled with linear complexity through selective scanning and state updating,providing a promising approach for balancing global representation learning and computational efficiency in remote sensing image super-resolution.This paper systematically reviews the theoretical foundations and method-ological framework of Mamba-based super-resolution,summarizes representative studies,and categorizes them into four major directions:frequency-domain modeling and spectral enhancement,structural fusion and attention enhancement,cross-modal and multi-source modeling,and lightweight design with re-param-eterization.Under a unified synthetic degradation evaluation protocol,representative methods are compar-atively analyzed in terms of reconstruction performance,perceptual quality,and applicability on bench-mark datasets,including AID,DIOR,and UCAS_AOD.The results demonstrate that Mamba-based methods exhibit substantial potential for preserving structural continuity,restoring geometric consistency,and suppressing pseudo-textures,while achieving competitive reconstruction performance across multiple datasets.Finally,key challenges and future trends,including real-world degradations,cross-domain gen-eralization,and reproducible benchmarks,are discussed,providing valuable references for future research and practical deployment.