Answers are grounded in this paper and cite their sources.
Pansharpening combines panchromatic (PAN) and multispectral (MS) images to recover high-resolution MS imagery. Many methods treat the two inputs as equivalent feature sources and fuse them by concatenation or addition. This treatment can transfer unwanted high-frequency content from the PAN image and disturb interband relationships in the MS features. We instead regard the PAN image as a spatial constraint and propose PSMamba, an asymmetric conditional guidance framework. Its dynamic spatial filtering mechanism uses Mamba's selective scan to retain useful PAN structures while limiting texture injection. Its conditional spectral calibration mechanism adjusts MS feature distributions under spatial guidance to preserve interband proportions. On the WorldView-II, WorldView-III, and Gaofen-2 datasets, PSMamba improves PSNR over Pan-Mamba by up to 0.89 dB and reduces SAM by up to 6.8%, indicating better spectral fidelity with a controlled spatial-spectral tradeoff.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: