FuBay: An Integrated Fusion Framework for Hyperspectral Super-Resolution Based on Bayesian Tensor Ring.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral image (HSI) spatial enhancement is crucial for detailed analysis.
• Low-rank tensor methods offer advantages but struggle with manual rank selection and parameter tuning.
• Existing approaches lack exploration of underlying low-dimensional factors.

Purpose of the Study:

• To introduce FuBay, a novel Bayesian sparse learning-based tensor ring (TR) fusion model for HSI spatial enhancement.
• To develop a fully Bayesian probabilistic tensor framework that addresses limitations of current fusion methods.
• To eliminate the need for manual parameter tuning in hyperspectral fusion.

Main Methods:

• Proposed a Bayesian sparse learning-based tensor ring (TR) fusion model (FuBay).
• Utilized hierarchical sparsity-inducing prior distributions for a fully Bayesian probabilistic approach.
• Implemented a component pruning mechanism to determine the true latent tensor rank.
• Derived a variational inference (VI)-based algorithm to learn the posterior of TR factors, avoiding non-convex optimization.

Main Results:

• FuBay demonstrated superior performance compared to state-of-the-art hyperspectral fusion methods in extensive experiments.
• The proposed component pruning effectively determined the latent tensor rank, addressing a key limitation of prior methods.
• The variational inference algorithm successfully learned tensor factors without encountering non-convex optimization issues.

Conclusions:

• FuBay offers a parameter-tuning-free solution for hyperspectral image spatial enhancement.
• The Bayesian probabilistic tensor framework provides a robust and effective approach to HSI fusion.
• This novel method advances the field of hyperspectral image processing by overcoming significant challenges in tensor-based fusion.