MATLAB implementation of the Beta Compositional Model algorithm for Hyperspectral unmixing. See related paper, doi: 10.1109/JSTARS.2014.2330347
***If this code is used, cite it: Xiaoxiao Du, & Alina Zare. (2019, April 12). GatorSense/BetaCompositionalModel: Initial Release (Version v1.0). Zenodo. http://doi.org/10.5281/zenodo.2638288
***NOTE: If the BCM Unmixing Algorithm is used in any publication or presentation, the following reference must be cited:
X. Du, A. Zare, P. Gader, D. Dranishnikov, ìSpatial and Spectral Unmixing Using the Beta Compositional Model,î IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 7, no. 6, pp. 1994-2003, June 2014.
The BCM Unmixing Algorithm runs using the following function:
[Parameters] = BCMParameters(endmembers);
[P] = BCM(Xim, Parameters, MethodFlag)
The endmembers input is a 1xM cell of endmember samples (known) where M is the number of endmembers.
DxNumESamples double matrix within each cell, where D is the number of spectral bands and NumESamples is the number of endmember samples.
The Xim input is a N1xN2xD matrix of N1xN2 image data points of dimensionality D.
The parameters input is a struct with the following fields:
Parameters - struct - The struct contains the following fields:
% These parameters are user defined (can be modified)
1. NumberIterations: Number of iterations
2. K: Number of nearest neighbors
3. c: Number of clusters for spatial K-means
4. SigV: Weighting on SigmaV
5. SigM: Weighting on SigmaMean
6. s: Scaling parameter for location information
7. ECovariance: diagonal covariance on endmembers, currently same for all endmembers
% These parameters are computed from input data and known endmembers (does not need modification)
8. MaxNumEMs, M: Number of endmembers
9. bandnum, D: Number of spectral bands/dimensions
10. N: Number of pixels(input data points)
11. Emean: Mean of endmember samples
12. BetaParameters: parameters for fitting a Beta distribution to each band of each endmember
*Parameters can be modified in [Parameters] = BCMParameters(endmembers) function.
The MethodFlag input is an 1x1 integer, choose from 1, 2, 3, 4, where each number indicates one BCM approach:
1. BCM-Spectral-QP (Quadratic Programming Approach)
2. BCM-Spectral-MH (Metropolis-Hastings Sampling Approach)
3. BCM-Spatial-QP
4. BCM-Spatial-MH
Description of files:
demo_allMethods.m - This is the main file. The demo performs BCM unmixing and compares with FCLS and NCM unmixing methods.
BCM.m - BCM (Beta Compositional Model) Spatial and Spectral Unmixing Algorithms.
BCMParameters.m - Sets the parameters to be used during the BCM unmixing algorithm. Please change accordingly.
BCM_Spectral.m- Performs BCM-Spectral unmixing, including BCM-Spectral-QP and BCM-Spectral-MH approaches.
BCM_Spatial.m - Performs BCM-Spatial unmixing, including BCM-Spatial-QP and BCM-Spatial-MH approaches.
hyperFcls.m - Performs fully constrained least squares on pixels of M. Written by I. Gerg (2012): Matlab Hyperspectral Toolbox [Online]. Available at https://github.com/isaacgerg/matlabHyperspectralToolbox.git.
unmix_qpas_correct.m - Performs quadratic unmixing given endmember mean values.
unmix2.m - Performs NCM-QP unmixing.
unmixGaussian.m - Performs NCM-MH unmixing.
PError.m - Computes the proportion errors per pixel per endmember given true and estimated proportion values.
demo.mat - Hyperspectral image and endmember samples (provided for demo purposes)
Ptrue.mat - Manual ground truth for the demo hyperspectral image (provided for demo purposes)
Authors: Xiaoxiao Du, Alina Zare
University of Missouri, Department of Electrical and Computer Engineering
Email Address: xdy74@mail.missouri.edu; zarea@missouri.edu
Latest Revision: November 3, 2014
This code uses MATLAB Statistics Toolbox and Matlab Optimization Toolbox.