Relevant and invariant feature selection of hyperspectral images for domain generalization

Claudio Persello, Lorenzo Bruzzone

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

5 Citations (Scopus)

Abstract

This paper presents a novel feature selection method for the analysis of hyperspectral images. The proposed method aims at selecting a subset of the original features that are both 1) relevant for the considered problem (i.e., preserve the functional relationship between input and output variables), and 2) invariant (stable) across different domains (i.e., minimize the data set shift among different domains). Domains can be associated with images collected on different areas or on the same area at different times. We propose a novel measure of domain stability, which evaluates the distance of the conditional distributions between the source and target domain. Such a measure is defined on the basis of kernel embeddings of conditional distributions and can be applied to both classification and regression problems. Experimental results show the effectiveness of the proposed method in selecting features with high generalization capabilities on the target domain.

Original languageEnglish
Title of host publicationInternational Geoscience and Remote Sensing Symposium (IGARSS)
Place of PublicationPiscataway, NJ
PublisherIEEE
Pages3562-3565
Number of pages4
ISBN (Electronic)978-1-4799-5775-0
DOIs
Publication statusPublished - 4 Nov 2014
EventJoint International Geoscience and Remote Sensing Symposium (IGARSS 2014) / 35th Canadian Symposium on Remote Sensing (35th CSRS): Energy and our Changing Planet - Quebec City Convention Centre, Quebec, Canada
Duration: 13 Jul 201418 Jul 2014

Publication series

NameInternational Geoscience and Remote Sensing Symposium (IGARSS)
PublisherIEEE
Volume2014
ISSN (Print)2153-6996
ISSN (Electronic)2153-7003

Conference

ConferenceJoint International Geoscience and Remote Sensing Symposium (IGARSS 2014) / 35th Canadian Symposium on Remote Sensing (35th CSRS)
Abbreviated titleIGARSS 2014
CountryCanada
CityQuebec
Period13/07/1418/07/14

Fingerprint

Feature extraction
method
distribution
analysis
preserve

Keywords

  • Feature selection
  • Hyperspectral data
  • Image classification
  • Kernel methods
  • Remote sensing

Cite this

Persello, C., & Bruzzone, L. (2014). Relevant and invariant feature selection of hyperspectral images for domain generalization. In International Geoscience and Remote Sensing Symposium (IGARSS) (pp. 3562-3565). [6947252] (International Geoscience and Remote Sensing Symposium (IGARSS); Vol. 2014). Piscataway, NJ: IEEE. https://doi.org/10.1109/IGARSS.2014.6947252
Persello, Claudio ; Bruzzone, Lorenzo. / Relevant and invariant feature selection of hyperspectral images for domain generalization. International Geoscience and Remote Sensing Symposium (IGARSS). Piscataway, NJ : IEEE, 2014. pp. 3562-3565 (International Geoscience and Remote Sensing Symposium (IGARSS)).
@inproceedings{db376ca8c48d40cb9c7212abd1db6e00,
title = "Relevant and invariant feature selection of hyperspectral images for domain generalization",
abstract = "This paper presents a novel feature selection method for the analysis of hyperspectral images. The proposed method aims at selecting a subset of the original features that are both 1) relevant for the considered problem (i.e., preserve the functional relationship between input and output variables), and 2) invariant (stable) across different domains (i.e., minimize the data set shift among different domains). Domains can be associated with images collected on different areas or on the same area at different times. We propose a novel measure of domain stability, which evaluates the distance of the conditional distributions between the source and target domain. Such a measure is defined on the basis of kernel embeddings of conditional distributions and can be applied to both classification and regression problems. Experimental results show the effectiveness of the proposed method in selecting features with high generalization capabilities on the target domain.",
keywords = "Feature selection, Hyperspectral data, Image classification, Kernel methods, Remote sensing",
author = "Claudio Persello and Lorenzo Bruzzone",
year = "2014",
month = "11",
day = "4",
doi = "10.1109/IGARSS.2014.6947252",
language = "English",
series = "International Geoscience and Remote Sensing Symposium (IGARSS)",
publisher = "IEEE",
pages = "3562--3565",
booktitle = "International Geoscience and Remote Sensing Symposium (IGARSS)",
address = "United States",

}

Persello, C & Bruzzone, L 2014, Relevant and invariant feature selection of hyperspectral images for domain generalization. in International Geoscience and Remote Sensing Symposium (IGARSS)., 6947252, International Geoscience and Remote Sensing Symposium (IGARSS), vol. 2014, IEEE, Piscataway, NJ, pp. 3562-3565, Joint International Geoscience and Remote Sensing Symposium (IGARSS 2014) / 35th Canadian Symposium on Remote Sensing (35th CSRS), Quebec, Canada, 13/07/14. https://doi.org/10.1109/IGARSS.2014.6947252

Relevant and invariant feature selection of hyperspectral images for domain generalization. / Persello, Claudio; Bruzzone, Lorenzo.

International Geoscience and Remote Sensing Symposium (IGARSS). Piscataway, NJ : IEEE, 2014. p. 3562-3565 6947252 (International Geoscience and Remote Sensing Symposium (IGARSS); Vol. 2014).

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

TY - GEN

T1 - Relevant and invariant feature selection of hyperspectral images for domain generalization

AU - Persello, Claudio

AU - Bruzzone, Lorenzo

PY - 2014/11/4

Y1 - 2014/11/4

N2 - This paper presents a novel feature selection method for the analysis of hyperspectral images. The proposed method aims at selecting a subset of the original features that are both 1) relevant for the considered problem (i.e., preserve the functional relationship between input and output variables), and 2) invariant (stable) across different domains (i.e., minimize the data set shift among different domains). Domains can be associated with images collected on different areas or on the same area at different times. We propose a novel measure of domain stability, which evaluates the distance of the conditional distributions between the source and target domain. Such a measure is defined on the basis of kernel embeddings of conditional distributions and can be applied to both classification and regression problems. Experimental results show the effectiveness of the proposed method in selecting features with high generalization capabilities on the target domain.

AB - This paper presents a novel feature selection method for the analysis of hyperspectral images. The proposed method aims at selecting a subset of the original features that are both 1) relevant for the considered problem (i.e., preserve the functional relationship between input and output variables), and 2) invariant (stable) across different domains (i.e., minimize the data set shift among different domains). Domains can be associated with images collected on different areas or on the same area at different times. We propose a novel measure of domain stability, which evaluates the distance of the conditional distributions between the source and target domain. Such a measure is defined on the basis of kernel embeddings of conditional distributions and can be applied to both classification and regression problems. Experimental results show the effectiveness of the proposed method in selecting features with high generalization capabilities on the target domain.

KW - Feature selection

KW - Hyperspectral data

KW - Image classification

KW - Kernel methods

KW - Remote sensing

UR - http://www.scopus.com/inward/record.url?scp=84911372070&partnerID=8YFLogxK

U2 - 10.1109/IGARSS.2014.6947252

DO - 10.1109/IGARSS.2014.6947252

M3 - Conference contribution

AN - SCOPUS:84911372070

T3 - International Geoscience and Remote Sensing Symposium (IGARSS)

SP - 3562

EP - 3565

BT - International Geoscience and Remote Sensing Symposium (IGARSS)

PB - IEEE

CY - Piscataway, NJ

ER -

Persello C, Bruzzone L. Relevant and invariant feature selection of hyperspectral images for domain generalization. In International Geoscience and Remote Sensing Symposium (IGARSS). Piscataway, NJ: IEEE. 2014. p. 3562-3565. 6947252. (International Geoscience and Remote Sensing Symposium (IGARSS)). https://doi.org/10.1109/IGARSS.2014.6947252