Superpixel construction for hyperspectral unmixing

Zeng Li, Jie Chen, Susanto Rahardja

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

15 Scopus citations

Abstract

Spectral unmixing aims to determine the component materials and their associated abundances from mixed pixels in a hyperspectral image. Instead of performing unmixing independently on each pixel, investigating spatial and spectral correlations among pixels can be beneficial to enhance the unmixing performance. However linking pixels across an entire image for such a purpose can be computationally cumbersome and physically unreasonable. In order to address this issue, we propose to construct superpixels for hyperspectral data unmixing. Using an SLIC-based (Simple Linear Iterative Clustering) superpixel constructing process, adjacent pixels are clustered into several blocks with similar spectral signatures. After this preprocessing, unmixing is then performed with a graph-based total variation regularization to benefit from the heterogeneity within each superpixel. Experimental results on synthetic data and real hyperspectral data illustrate advantages of the proposed scheme.

Original languageEnglish
Title of host publication2018 26th European Signal Processing Conference, EUSIPCO 2018
PublisherEuropean Signal Processing Conference, EUSIPCO
Pages647-651
Number of pages5
ISBN (Electronic)9789082797015
DOIs
StatePublished - 29 Nov 2018
Event26th European Signal Processing Conference, EUSIPCO 2018 - Rome, Italy
Duration: 3 Sep 20187 Sep 2018

Publication series

NameEuropean Signal Processing Conference
Volume2018-September
ISSN (Print)2219-5491

Conference

Conference26th European Signal Processing Conference, EUSIPCO 2018
Country/TerritoryItaly
CityRome
Period3/09/187/09/18

Keywords

  • Graph regularization
  • Hyperspectral images
  • Spectral unmixing
  • Superpixel analysis

Fingerprint

Dive into the research topics of 'Superpixel construction for hyperspectral unmixing'. Together they form a unique fingerprint.

Cite this