Skip to content

etienne-monier/lib-unmixing

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 

Repository files navigation

A Python3 library for basic unmixing functions

Functions list

The three functions implemented in this library are:

  • Vertex Component Analysis (VCA). Related article is J. Nascimento and J. Dias, "Vertex Component Analysis: A fast algorithm to unmix hyperspectral data", IEEE Transactions on Geoscience and Remote Sensing, vol. 43, no. 4, pp. 898-910, 2005.
  • Simplex Identification via Split Augmented Lagrangian (SISAL). Related article is J. Bioucas-Dias, "A variable splitting augmented Lagrangian approach to linear spectral unmixing", in First IEEE GRSS Workshop on Hyperspectral Image and Signal Processing-WHISPERS'2009, Grenoble, France, 2009.
  • Sparse Unmixing via variable Splitting and Augmented Lagrangian methods(SUNSAL). Related article is Bioucas-Dias, J. M., & Figueiredo, M. A., "Alternating direction algorithms for constrained sparse regression: Application to hyperspectral unmixing." in Workshop in Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), 2010 (pp. 1-4).

Matlab versions of these codes are available in the Jose Bioucas Dias website.

VCA function

This function were translated by Adrien Lagrange (view his github page).

### Usage

Ae, indice, Yp = vca(Y,R,verbose = True,snr_input = 0)

Input variables

Y - matrix with dimensions L(channels) x N(pixels) each pixel is a linear mixture of R endmembers signatures Y = M x s, where

  • s = gamma x alpha
  • gamma is a illumination perturbation factor and
  • alpha are the abundance fractions of each endmember.

R - positive integer number of endmembers in the scene

Output variables

Ae - estimated mixing matrix (endmembers signatures)

indice - pixels that were chosen to be the most pure

Yp - Data matrix Y projected.

Optional parameters

snr_input - (float) signal to noise ratio (dB)

v - [True | False]

Author, license and info

Author: Adrien Lagrange ([email protected])

This code is a translation of a matlab code provided by Jose Nascimento ([email protected]) and Jose Bioucas Dias ([email protected]) available at http://www.lx.it.pt/~bioucas/code.htm under the GNU General Public License 2.0.

Translation of last version at 22-February-2018 (Matlab version 2.1 (7-May-2004)).

SISAL function

Usage

M,Up,my,sing_values = sisal(Y,p,**kwargs)

### Description

Simplex identification via split augmented Lagrangian (SISAL) estimates the vertices M={m_1,...m_p} of the (p-1)-dimensional simplex of minimum volume containing the vectors [y_1,...y_N], under the assumption that y_i belongs to a (p-1) dimensional affine set.

For details see José M. Bioucas-Dias, "A variable splitting augmented lagrangian approach to linear spectral unmixing", First IEEE GRSS Workshop on Hyperspectral Image and Signal Processing - WHISPERS, 2009. (http://arxiv.org/abs/0904.4635v1)

Input

Y - matrix with dimension L(channels) x N(pixels). Each pixel is a linear mixture of p endmembers signatures Y = M*x + noise.

p - number of independent columns of M. Therefore, M spans a (p-1)-dimensional affine set. p is the number of endmembers.

Optional input

mm_iters - Maximum number of constrained quadratic programs. Default: 80

tau - Regularization parameter in the problem

             Q^* = arg min_Q  -\log abs(det(Q)) + tau*|| Q*yp ||_h
                   subject to np.ones((1,p))*Q=mq
             where mq = ones(1,N)*yp'inv(yp*yp) and ||x||_h is the "hinge" induced norm.
   Default: 1

mu - Augmented Lagrange regularization parameter. Default: 1

spherize - {True, False} Applies a spherization step to data such that the spherized data spans over the same range along any axis. Default: True

tolf - Tolerance for the termination test (relative variation of f(Q)). Default: 1e-2

M0 - Initial M, dimension L x p. Defaults is given by the VCA algorithm.

verbose - {0,1,2,3}

  • 0 - work silently
  • 1 - display simplex volume
  • 2 - display figures
  • 3 - display SISAL information
  • 4 - display SISAL information and figures Default: 1

Output

M - estimated endmember signature matrix L x p

Up - isometric matrix spanning the same subspace as M, imension is L x p

my - mean value of Y

sing_values - (p-1) eigenvalues of Cy = (y-my)*(y-my)/N. The dynamic range of these eigenvalues gives an idea of the difficulty of the underlying problem

Note

The identified affine set is given by

{z\in R^p : z=Up(:,1:p-1)*a+my, a\in R^(p-1)}

Author, license and info

Author: Etienne Monier ([email protected])

This code is a translation of a matlab code provided by Jose Nascimento ([email protected]) and Jose Bioucas Dias ([email protected]) available at http://www.lx.it.pt/~bioucas/code.htm under the GNU General Public License 2.0.

Translation of last version at 20-April-2018 (Matlab version 2.1 (7-May-2004))

SUNSAL function

Usage

x = sunsal_v2(M,Y,**kwargs)

Description

SUNSAL (sparse unmixing via variable splitting and augmented Lagrangian methods) algorithm implementation. Accepted constraints are:

    1. Positivity: X >= 0
    1. Addone: np.sum(X,axis=0) = np.ones(N)

For details see J. Bioucas-Dias and M. Figueiredo, “Alternating direction algorithms for constrained sparse regression: Application to hyperspectral unmixing”, in 2nd IEEE GRSS Workshop on Hyperspectral Image and Signal Processing-WHISPERS'2010, Raykjavik, Iceland, 2010.

Input

M - endmember signature matrix with dimensions L(channels) x p(endmembers)

Y - matrix with dimensions L(channels) x N(pixels). Each pixel is a linear mixture of p endmembers signatures

Optional input

al_iters - Minimum number of augmented Lagrangian iterations. Default: 100

lambda_p - regularization parameter. lambda is either a scalar or a vector with N components (one per column of x). Default: 0

positivity - {True, False} Enforces the positivity constraint. Default: False

addone - {True, False} Enforces the addone constraint. Default: False

tol - tolerance for the primal and dual residuals. Default: 1e-4

verbose = {True, False}

  • False - work silently
  • True - display iteration info Default: True

Output

X - estimated abundance matrix of size p x N

Author, license and info

Author: Etienne Monier ([email protected])

This code is a translation of a matlab code provided by Jose Nascimento ([email protected]) and Jose Bioucas Dias ([email protected]) available at http://www.lx.it.pt/~bioucas/code.htm under the GNU General Public License 2.0.

Translation of last version at 20-April-2018 (Matlab version 2.1 (7-May-2004))

Authors

Software translated from matlab to python by Etienne Monier ([email protected]), 2018.

Initial matlab author: Jose Bioucas-Dias, 2009

License

This code is distributed under the terms of the GNU General Public License 2.0.

About

Python3 library for common unmixing functions

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages