Parameter Estimation for Blind l<sub>q</sub> Hyperspectral Unmixing Using Bayesian Optimization

Jakob Sigurdsson, Magnus O. Ulfarsson, Johannes R. Sveinsson

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Blind hyperspectral unmixing is the task of estimating both the pure material spectra (endmembers), and the abundances in hyperspectral images. The performance of hyperspectral unmixing methods is very often dependent on tuning parameters. Accurately estimating these parameters is computationally intensive and this can severely limit the complexity of the underlying model. In this paper, we propose using Bayesian optimization to estimate tuning parameters for blind hyperspectral unmixing. Using real data, we show that the proposed method can be successfully applied to estimate tuning parameters for hyperspectral unmixing. Also, we show that increasing the number of tuning parameters can improve the unmixing results.

Original languageEnglish
Title of host publicationWorkshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing
Subtitle of host publicationEvolution in Remote Sensing, WHISPERS 2018
PublisherIEEE Computer Society
ISBN (Electronic)9781728115818
DOIs
Publication statusPublished - Sept 2018
Event9th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing, WHISPERS 2018 - Amsterdam, Netherlands
Duration: 23 Sept 201826 Sept 2018

Publication series

NameWorkshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing
Volume2018-September

Conference

Conference9th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing, WHISPERS 2018
Country/TerritoryNetherlands
CityAmsterdam
Period23/09/1826/09/18

Bibliographical note

Funding Information: This work was supported by the Postdoctoral Fund of the University of Iceland and the Research Fund of the University of Iceland Publisher Copyright: © 2018 IEEE.

Other keywords

  • Bayesian optimization
  • Hyperspectral unmixing
  • lq regularization
  • parameter estimation

Fingerprint

Dive into the research topics of 'Parameter Estimation for Blind l<sub>q</sub> Hyperspectral Unmixing Using Bayesian Optimization'. Together they form a unique fingerprint.

Cite this