Learning the relevant image features with multiple kernels

Devis Tuia*, Giona Matasci, Gustavo Camps-Valls, Mikhail Kanevski

*Corresponding author for this work

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

2 Citations (Scopus)

Abstract

This paper proposes to learn the relevant features of remote sensing images for automatic spatio-spectral classification with the automatic optimization of multiple kernels. The method consists of building dedicated kernels for different sets of bands, contextual or textural features. The optimal linear combination of kernels is optimized through gradient descent on the support vector machine (SVM) objective function. Since a naïve implementation is computationally demanding, we propose an efficient model selection procedure based on kernel alignment. The result is a weight -learned from the data- for each kernel where both relevant and meaningless image features emerge after training. Excellent results are observed in both multi and hyperspectral image classification, improving standard SVM and other spatio-spectral formulations.

Original languageEnglish
Title of host publication2009 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2009 - Proceedings
PagesII65-II68
DOIs
Publication statusPublished - Dec 2009
Externally publishedYes
Event2009 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2009 - Cape Town, South Africa
Duration: 12 Jul 200917 Jul 2009

Publication series

NameInternational Geoscience and Remote Sensing Symposium (IGARSS)
Volume2

Conference

Conference2009 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2009
CountrySouth Africa
CityCape Town
Period12/07/0917/07/09

Keywords

  • Image classification
  • Kernel alignment
  • Multiple kernel learning (MKL)
  • SimpleMKL
  • Support vector machine (SVM)

Fingerprint Dive into the research topics of 'Learning the relevant image features with multiple kernels'. Together they form a unique fingerprint.

Cite this