A multi-label Hyperspectral image classification method with deep learning features

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

12 Scopus citations

Abstract

Hyperspectral image (HSI) classification is an important application of HSI analysis, which aims at assigning a class label to each pixel. However, considering that mixed pixels commonly exist in HSI, assigning a unique label to each pixel is imprecise. To better analysis the scene imaged in an HSI, we propose a multi-label hyperspectral image classification approach based on deep learning in this study. First, stacked denoising autoencoder (SDAE) method is used to extract deep features for each pixel without supervision, which can well represent the nonlinearity of the mixed pixels in a high dimensional feature space. Then, multi-label logistic regression method assigns each pixel multi labels. Experimental results on the synthetic data, real hyperspectral data and down-sampling hyperspectral data demonstrate the effectiveness of the proposed method.

Original languageEnglish
Title of host publicationProceedings of the International Conference on Internet Multimedia Computing and Service, ICIMCS 2016
PublisherAssociation for Computing Machinery
Pages127-131
Number of pages5
ISBN (Electronic)9781450348508
DOIs
StatePublished - 19 Aug 2016
Event8th International Conference on Internet Multimedia Computing and Service, ICIMCS 2016 - Xi'an, China
Duration: 19 Aug 201621 Aug 2016

Publication series

NameACM International Conference Proceeding Series
Volume19-21-August-2016

Conference

Conference8th International Conference on Internet Multimedia Computing and Service, ICIMCS 2016
Country/TerritoryChina
CityXi'an
Period19/08/1621/08/16

Keywords

  • Hyperspectral image
  • Logistic regression
  • Multi-label classification
  • Stacked denoising autoencoder

Fingerprint

Dive into the research topics of 'A multi-label Hyperspectral image classification method with deep learning features'. Together they form a unique fingerprint.

Cite this