Home / Research /

Hyperspectral unmixing

Our activities deal with spectral unmixing of hyperspectral data.

For these models...

• LMM : Linear Mixing Model
     (the most popular model)
• GBM : Generalized Bilinear Model
     (a nonlinear model to address multiple scattering effects)
• PPNMM : Polynomial Post-Nonlinear Mixing Model
     (a flexible nonlinear model)
• RLMM : Robust Linear Mixing Model
     (a group-sparse model generalizing the LMM)
• NCM : Normal Compositional Model
     (statistical model to address spectral variability)
• PLMM : Perturbed Linear Mixing Model
     (deterministic model to address spectral variability)

under these scenarios...

• supervised : inversion step (spectral signatures are known)
• unsupervised : joint estimation of spectral signatures and abundances

and these contexts

• online (on-the-fly processing)
• off-line (batch processing)
• partially asynchronous distributed processing

using these methods...

• Bayesian estimation and MCMC sampling
• (scaled) gradient optimization
• block coordinate descent algorithm (with majorization-minimization steps)
• ADMM algorithm
• PALM algorithm

The considered applicative context mainly deals with remote sensing. However, the proposed algorithms/methodologies have been also applied for

• spectrochemical analysis : .
• gene expression factor analysis : .
• electron energy-loss spectroscopy (EELS) analysis : .

A review paper on spectral unmixing has been published in IEEE J. Sel. Topics Topics Applied Earth Observations and Remote Sensing in 2012:

• article [ .pdf - 3.7Mo ].

A tutorial paper on nonlinear unmixing has been published in IEEE Signal Processing Magazine in 2014:

• article [ .pdf - 3.8Mo ].

A preliminary study to validate nonlinear unmixing models in vegetated areas has been published in the IEEE J. Sel. Topics Appl. Earth Observations Remote Sensing. More information is available here:

• nonlinear mixing model validation .

MCMC algorithm for unsupervised LMM-based unmixing: .

New! A beta version of a Matlab GUI for the Bayesian Linear Unmixing (BLU) algorithm:

• Matlab GUI BLU code [ .zip - 2.2Mo ]

MCMC algorithm for supervised LMM-based unmixing: .

MCMC algorithm for supervised NCM-based unmixing: .

MCMC algorithm for unsupervised NCM-based unmixing: .

Gradient algorithm for supervised LMM-based unmixing: .

Robust NMF for unsupervised RLMM-based unmixing: .

Weighted NMF for unsupervised unmixing: .

Multiscale regularization for fast sparse unmixing: .

ADMM for unsupervised PLMM-based unmixing: .

Online PLMM-based unmixing of multi-temporal images: .

Partially asynchronous distributed unmixing: .

Gradient algorithm for supervised GBM-based unmixing: .

Gradient algorithm for supervised PPNMM-based unmixing: .

Detection of non-linear mixtures using PPNMM-based unmixing: .

Detection of non-linear mixtures using subspace estimation: .