Collaborative sparse regression using spatially correlated supports--Application to hyperspectral unmixing.

Area of Science:

• Remote Sensing
• Computer Vision
• Statistical Modeling

Background:

• Hyperspectral imaging provides rich spectral information for material identification.
• Linear unmixing aims to identify constituent materials and their abundances in mixed pixels.
• Existing methods often struggle with spatial correlations and parameter selection.

Purpose of the Study:

• To develop a novel Bayesian collaborative sparse regression method for hyperspectral image linear unmixing.
• To incorporate spatial correlations of material abundances across pixels.
• To automate regularization parameter selection.

Main Methods:

• A Bayesian model for structured sparse regression using a truncated multivariate Ising Markov random field.
• Incorporation of pixel non-emptiness and material-specific spatial regularity priors.
• An advanced Markov chain Monte Carlo (MCMC) algorithm for posterior estimation and abundance calculation.
• Self-adjusting regularization parameter estimation within the MCMC algorithm.

Main Results:

• The proposed method effectively models spatially correlated material distributions.
• The MCMC algorithm accurately estimates posterior probabilities and abundance vectors.
• The self-adjusting parameter estimation eliminates the need for cross-validation.
• Experimental results demonstrate superior performance compared to existing algorithms on synthetic and real data.

Conclusions:

• The novel Bayesian approach enhances hyperspectral image linear unmixing by leveraging spatial information.
• The automated parameter tuning simplifies the application of the method.
• This work offers a robust and efficient solution for analyzing hyperspectral data.