Work

Physics-Consistent Unsupervised Fusion of PRISMA and Sentinel-2 Images

Year: 2026
Type: conference-paper
Abstract: Hyperspectral–multispectral image fusion aims to reconstruct high-resolution hyperspectral images by combining the complementary spatial and spectral information from different sensors. In practical satellite missions such as PRISMA and Sentinel-2, the absence of pixel-wise high-resolution hyperspectral ground truth makes supervised learning and conventional full-reference evaluation infeasible. In this paper, we propose PCUF, a physics-consistent unsupervised fusion framework that integrates observation consistency constraints with learning-based reconstruction. The proposed method enforces spectral–spatial consistency between PRISMA and Sentinel-2 observations without relying on ground-truth hyperspectral images. To enable reliable performance assessment under this no-ground-truth setting, in-situ farmland spectral measurements are employed as external references, and spectral fidelity is quantitatively evaluated using the Spectral Angle Mapper (SAM). Experimental results on real PRISMA and Sentinel-2 data demonstrate that the proposed method achieves improved spectral consistency compared with representative fusion approaches, while preserving fine spatial structures. These results demonstrate the effectiveness of physics-consistent constraints for unsupervised hyperspectral–multispectral fusion in real-world scenarios.
Language: English
Open Access status: hybrid
OpenAlex ID: W7223100240
FWCI:
Cites: 20
Cited by:
Related to: 10