Regularization parameter selection in minimum volume hyperspectral unmixing

Lina Zhuang, Chia Hsiang Lin, Mario A.T. Figueiredo, Jose M. Bioucas-Dias

Research output: Contribution to journalArticle

Abstract

Linear hyperspectral unmixing (HU) aims at factoring the observation matrix into an endmember matrix and an abundance matrix. Linear HU via variational minimum volume (MV) regularization has recently received considerable attention in the remote sensing and machine learning areas, mainly owing to its robustness against the absence of pure pixels. We put some popular linear HU formulations under a unifying framework, which involves a data-fitting term and an MV-based regularization term, and collectively solve it via a nonconvex optimization. As the former and the latter terms tend, respectively, to expand (reducing the data-fitting errors) and to shrink the simplex enclosing the measured spectra, it is critical to strike a balance between those two terms. To the best of our knowledge, the existing methods find such balance by tuning a regularization parameter manually, which has little value in unsupervised scenarios. In this paper, we aim at selecting the regularization parameter automatically by exploiting the fact that a too large parameter overshrinks the volume of the simplex defined by the endmembers, making many data points be left outside of the simplex and hence inducing a large data-fitting error, while a sufficiently small parameter yields a large simplex making data-fitting error very small. Roughly speaking, the transition point happens when the simplex still encloses the data cloud but there are data points on all its facets. These observations are systematically formulated to find the transition point that, in turn, yields a good parameter. The competitiveness of the proposed selection criterion is illustrated with simulated and real data.

Original languageEnglish
Article number8798985
Pages (from-to)9858-9877
Number of pages20
JournalIEEE Transactions on Geoscience and Remote Sensing
Volume57
Issue number12
DOIs
Publication statusPublished - 2019 Dec

Fingerprint

Learning systems
Remote sensing
Tuning
Pixels
matrix
parameter
competitiveness
pixel
remote sensing
machine learning
method

All Science Journal Classification (ASJC) codes

  • Electrical and Electronic Engineering
  • Earth and Planetary Sciences(all)

Cite this

Zhuang, Lina ; Lin, Chia Hsiang ; Figueiredo, Mario A.T. ; Bioucas-Dias, Jose M. / Regularization parameter selection in minimum volume hyperspectral unmixing. In: IEEE Transactions on Geoscience and Remote Sensing. 2019 ; Vol. 57, No. 12. pp. 9858-9877.
@article{fb4461fc835e44a3a6775a76f7aeeb78,
title = "Regularization parameter selection in minimum volume hyperspectral unmixing",
abstract = "Linear hyperspectral unmixing (HU) aims at factoring the observation matrix into an endmember matrix and an abundance matrix. Linear HU via variational minimum volume (MV) regularization has recently received considerable attention in the remote sensing and machine learning areas, mainly owing to its robustness against the absence of pure pixels. We put some popular linear HU formulations under a unifying framework, which involves a data-fitting term and an MV-based regularization term, and collectively solve it via a nonconvex optimization. As the former and the latter terms tend, respectively, to expand (reducing the data-fitting errors) and to shrink the simplex enclosing the measured spectra, it is critical to strike a balance between those two terms. To the best of our knowledge, the existing methods find such balance by tuning a regularization parameter manually, which has little value in unsupervised scenarios. In this paper, we aim at selecting the regularization parameter automatically by exploiting the fact that a too large parameter overshrinks the volume of the simplex defined by the endmembers, making many data points be left outside of the simplex and hence inducing a large data-fitting error, while a sufficiently small parameter yields a large simplex making data-fitting error very small. Roughly speaking, the transition point happens when the simplex still encloses the data cloud but there are data points on all its facets. These observations are systematically formulated to find the transition point that, in turn, yields a good parameter. The competitiveness of the proposed selection criterion is illustrated with simulated and real data.",
author = "Lina Zhuang and Lin, {Chia Hsiang} and Figueiredo, {Mario A.T.} and Bioucas-Dias, {Jose M.}",
year = "2019",
month = "12",
doi = "10.1109/TGRS.2019.2929776",
language = "English",
volume = "57",
pages = "9858--9877",
journal = "IEEE Transactions on Geoscience and Remote Sensing",
issn = "0196-2892",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "12",

}

Regularization parameter selection in minimum volume hyperspectral unmixing. / Zhuang, Lina; Lin, Chia Hsiang; Figueiredo, Mario A.T.; Bioucas-Dias, Jose M.

In: IEEE Transactions on Geoscience and Remote Sensing, Vol. 57, No. 12, 8798985, 12.2019, p. 9858-9877.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Regularization parameter selection in minimum volume hyperspectral unmixing

AU - Zhuang, Lina

AU - Lin, Chia Hsiang

AU - Figueiredo, Mario A.T.

AU - Bioucas-Dias, Jose M.

PY - 2019/12

Y1 - 2019/12

N2 - Linear hyperspectral unmixing (HU) aims at factoring the observation matrix into an endmember matrix and an abundance matrix. Linear HU via variational minimum volume (MV) regularization has recently received considerable attention in the remote sensing and machine learning areas, mainly owing to its robustness against the absence of pure pixels. We put some popular linear HU formulations under a unifying framework, which involves a data-fitting term and an MV-based regularization term, and collectively solve it via a nonconvex optimization. As the former and the latter terms tend, respectively, to expand (reducing the data-fitting errors) and to shrink the simplex enclosing the measured spectra, it is critical to strike a balance between those two terms. To the best of our knowledge, the existing methods find such balance by tuning a regularization parameter manually, which has little value in unsupervised scenarios. In this paper, we aim at selecting the regularization parameter automatically by exploiting the fact that a too large parameter overshrinks the volume of the simplex defined by the endmembers, making many data points be left outside of the simplex and hence inducing a large data-fitting error, while a sufficiently small parameter yields a large simplex making data-fitting error very small. Roughly speaking, the transition point happens when the simplex still encloses the data cloud but there are data points on all its facets. These observations are systematically formulated to find the transition point that, in turn, yields a good parameter. The competitiveness of the proposed selection criterion is illustrated with simulated and real data.

AB - Linear hyperspectral unmixing (HU) aims at factoring the observation matrix into an endmember matrix and an abundance matrix. Linear HU via variational minimum volume (MV) regularization has recently received considerable attention in the remote sensing and machine learning areas, mainly owing to its robustness against the absence of pure pixels. We put some popular linear HU formulations under a unifying framework, which involves a data-fitting term and an MV-based regularization term, and collectively solve it via a nonconvex optimization. As the former and the latter terms tend, respectively, to expand (reducing the data-fitting errors) and to shrink the simplex enclosing the measured spectra, it is critical to strike a balance between those two terms. To the best of our knowledge, the existing methods find such balance by tuning a regularization parameter manually, which has little value in unsupervised scenarios. In this paper, we aim at selecting the regularization parameter automatically by exploiting the fact that a too large parameter overshrinks the volume of the simplex defined by the endmembers, making many data points be left outside of the simplex and hence inducing a large data-fitting error, while a sufficiently small parameter yields a large simplex making data-fitting error very small. Roughly speaking, the transition point happens when the simplex still encloses the data cloud but there are data points on all its facets. These observations are systematically formulated to find the transition point that, in turn, yields a good parameter. The competitiveness of the proposed selection criterion is illustrated with simulated and real data.

UR - http://www.scopus.com/inward/record.url?scp=85075667290&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85075667290&partnerID=8YFLogxK

U2 - 10.1109/TGRS.2019.2929776

DO - 10.1109/TGRS.2019.2929776

M3 - Article

AN - SCOPUS:85075667290

VL - 57

SP - 9858

EP - 9877

JO - IEEE Transactions on Geoscience and Remote Sensing

JF - IEEE Transactions on Geoscience and Remote Sensing

SN - 0196-2892

IS - 12

M1 - 8798985

ER -