A Fast Multiscale Spatial Regularization for Sparse Hyperspectral Unmixing

This is the authors' implementation of [1].

The code is implemented in MATLAB and includes:

• demo1_MUA.m - a demo script comparing the algorithms (DC1)
• demo2_MUA.m - a demo script comparing the algorithms (DC2)
• demo3_MUA.m - a demo script comparing the algorithms (DC3)
• demo4_supp_MUA.m - demo script of the supplemental material
• demo5_supp_MUA.m - demo script of the supplemental material
• demo6_supp_MUA.m - demo script of the supplemental material
• sort_library_by_angle.m - sort the signatures in the spectral library
• prune_library.m - remove correlated signatures from the library
• tight_subplot.m - more efficient subplot
• soft.m - soft thresholding operator function
• sunsal.m - the SUnSAL algorithm
• sunsal_tv.m - the SUnSAL-TV algorithm
• sunsal_tv_lw_sp.m - the S2WSU algorithm
• sunsal_spreg.m - sparse unmixing at the fine spatial scale
• ./real_data/ - real images and spectral libraries used in the supplemental experiments
• ./vlfeat-0.9/ - the VLFeat toolbox (used for the SLIC superpixels alg.)
• ./HSI_segmentation/ - Binary partition tree HSI segmentation algorithm
• README - this file

IMPORTANT:

If you use this software please cite the following in any resulting publication:

[1] A Fast Multiscale Spatial Regularization for Sparse Hyperspectral Unmixing
    R.A. Borsoi, T. Imbiriba, J.C.M. Bermudez, C. Richard.
    IEEE Geoscience and Remote Sensing Letters, 2018.

INSTALLING & RUNNING:

Just start MATLAB and run demo1_MUA.m, demo2_MUA.m or demo3_MUA.m.

For the demos in the supplemental material, just run demo4_supp_MUA.m, demo5_supp_MUA.m, or demo6_supp_MUA.m. The data (spectral libraries and images) for these examples is included in the "real_data" folder.

The Cuprite HS image is not included, and should be downloaded from http://www.lx.it.pt/~bioucas/code.htm.

Important Notice About VLFeat

If you encounter problems with the "vl_slic" or "vl_setup" functions, try to download the latest version of the toolbox at http://www.vlfeat.org/install-matlab.html.

NOTES:

1. The SUnSAL and SUnSAL-TV algorithms are provided by Jose Bioucas Dias at http://www.lx.it.pt/~bioucas/code.htm.

2. The S2WSU algorithm was provided by Shaoquan Zhang, and corresponds to the publication: S. Zhang, J. Li, H.-C. Li, C. Deng and A. Plaza. Spectral-Spatial Weighted Sparse Regression for Hyperspectral Image Unmixing. IEEE Transactions on Geoscience and Remote Sensing, 2018

3. The vl_feat toolbox is included for the SLIC algorithm implementation http://www.vlfeat.org/

4. The Binary Partition Tree HSI segmentation algorithm was provided by Miguel Veganzones. Veganzones, M. A., Tochon, G., Dalla-Mura, M., Plaza, A. J., & Chanussot, J. Hyperspectral image segmentation using a new spectral unmixing-based binary partition tree representation. IEEE Transactions on Image Processing, 2014.