Patch normalizing flow regularization for hyperspectral pansharpening
Résumé
This paper presents an unsupervised neural network-based framework for fusing hyperspectral (HS) and multispectral (MS) images, addressing their inherent resolution trade-offs. Unlike supervised HS-MS fusion methods that require large training datasets, our approach is model-based and fully unsupervised. It is based on Principal Component Analysis (PCA) for spectral subspace identification and on an innovative Patch Normalizing Flow (Patch-NF) for spatial regularization. Experiments show that the proposed method offers a nice trade-off in terms of performance and computation time, when compared to alternative unsupervised baselines from the literature.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|