Dual-Domain Aligned Temporal–Spatial–Spectral Fusion Networks for No-Paired Hyperspectral and Multispectral Images

The objective of temporal–spatial–spectral fusion for remote sensing images is to reconstruct imagery with enhanced temporal, spatial, and spectral resolution. This process aims to integrate the rich spectral information inherent in hyperspectral images (HSIs) with the temporal and spatial details provided by multispectral images (MSIs). However, existing methodologies predominantly depend on the availability of strictly co-registered HSI-MSI pairs, constraining their practical applicability due to the significant challenge of simultaneously acquiring such paired data over the same geographical region. Furthermore, these methods frequently encounter difficulties in effectively harmonizing the inherent radiometric and spatial discrepancies between multisource remote sensing images, often resulting in suboptimal fusion results. To address these limitations, we propose a novel dual-domain aligned temporal–spatial–spectral fusion network (DAFN) designed for unpaired hyperspectral image (HSI) and multispectral image (MSI) imagery. DAFN integrates a style transfer module to mitigate radiometric inconsistencies, a spatial alignment module to rectify geometric misregistrations, dedicated modules for change feature extraction and spatial information refinement, culminating in feature fusion to accurately reconstruct high temporal–spatial resolution hyperspectral imagery. Extensive experiments on both simulated and real-world datasets demonstrate that DAFN consistently outperforms state-of-the-art methods.