Spectral unmixing for hyperspectral image classification with an adaptive endmember selection

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

2 Scopus citations

Abstract

Hyperspectral classification techniques are widely used for detailed analysis of the earth surface. However, mixed pixels caused by the relatively low spatial resolution of the imaging system are the big burden for traditional pure-pixel-hypothesis based hard classification methods. To address this problem, a novel method, which jointly uses soft classification and spectral unmixing, is proposed in this paper. The confusion matrix is exploited to determine the endmember set for each class. Then the generated endmember is adopted for spectral unmixing. The fractional abundance of training samples, which is generated from spectral unmixing, is utilized to optimize soft multinomial logistic regression classifier. The result of the optimized classifier will result in a more accurate confusion matrix. Thus, this procedure is executed iteratively to achieve required performance. Experimental results on synthetic and real hyperspectral data sets demonstrate the superiority of the proposed method for hyperspectral image classification.

Original languageEnglish
Title of host publicationIntelligence Science and Big Data Engineering - 4th International Conference, IScIDE 2013, Revised Selected Papers
PublisherSpringer Verlag
Pages359-367
Number of pages9
ISBN (Print)9783642420566
DOIs
StatePublished - 2013
Event4th International Conference on Intelligence Science and Big Data Engineering, IScIDE 2013 - Beijing, China
Duration: 31 Jul 20132 Aug 2013

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume8261 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference4th International Conference on Intelligence Science and Big Data Engineering, IScIDE 2013
Country/TerritoryChina
CityBeijing
Period31/07/132/08/13

Keywords

  • endmember selection
  • hyperspectral image
  • spectral unmixing
  • supervised classification

Fingerprint

Dive into the research topics of 'Spectral unmixing for hyperspectral image classification with an adaptive endmember selection'. Together they form a unique fingerprint.

Cite this