Weighted sparse graph based dimensionality reduction for hyperspectral images

Wei He, Hongyan Zhang, Liangpei Zhang, Wilfried Philips, Wenzhi Liao, Alejandro Frery (Editor)

Research output: Contribution to journalArticle

21 Citations (Scopus)

Abstract

Dimensionality reduction (DR) is an important and helpful preprocessing step for hyperspectral image (HSI) classification. Recently, sparse graph embedding (SGE) has been widely used in the DR of HSIs. In this letter, we propose a weighted sparse graph based DR (WSGDR) method for HSIs. Instead of only exploring the locality structure (as in neighborhood preserving embedding) or the linearity structure (as in SGE) of the HSI data, the proposed method couples the locality and linearity properties of HSI data together in a unified framework for the DR of HSIs. The proposed method was tested on two widely used HSI data sets, and the results suggest that the locality and linearity are complementary properties for HSIs. In addition, the experimental results also confirm the superiority of the proposed WSGDR method over the other state-of-the-art DR methods.
LanguageEnglish
Pages686-690
Number of pages5
JournalIEEE Geoscience and Remote Sensing Letters
Volume13
Issue number5
DOIs
Publication statusPublished - 18 Mar 2016

Fingerprint

linearity
Image classification
image classification
method

Keywords

  • sparse graph embedding
  • dimensionality reduction
  • nearest neighbor graph
  • hyperspectral image
  • weighted sparse coding
  • discriminant analysis
  • feature extraction

Cite this

He, Wei ; Zhang, Hongyan ; Zhang, Liangpei ; Philips, Wilfried ; Liao, Wenzhi ; Frery, Alejandro (Editor). / Weighted sparse graph based dimensionality reduction for hyperspectral images. In: IEEE Geoscience and Remote Sensing Letters. 2016 ; Vol. 13, No. 5. pp. 686-690.
@article{ed7ee588f12c4a41b0aff520290868cc,
title = "Weighted sparse graph based dimensionality reduction for hyperspectral images",
abstract = "Dimensionality reduction (DR) is an important and helpful preprocessing step for hyperspectral image (HSI) classification. Recently, sparse graph embedding (SGE) has been widely used in the DR of HSIs. In this letter, we propose a weighted sparse graph based DR (WSGDR) method for HSIs. Instead of only exploring the locality structure (as in neighborhood preserving embedding) or the linearity structure (as in SGE) of the HSI data, the proposed method couples the locality and linearity properties of HSI data together in a unified framework for the DR of HSIs. The proposed method was tested on two widely used HSI data sets, and the results suggest that the locality and linearity are complementary properties for HSIs. In addition, the experimental results also confirm the superiority of the proposed WSGDR method over the other state-of-the-art DR methods.",
keywords = "sparse graph embedding, dimensionality reduction, nearest neighbor graph, hyperspectral image, weighted sparse coding, discriminant analysis, feature extraction",
author = "Wei He and Hongyan Zhang and Liangpei Zhang and Wilfried Philips and Wenzhi Liao and Alejandro Frery",
year = "2016",
month = "3",
day = "18",
doi = "10.1109/LGRS.2016.2536658",
language = "English",
volume = "13",
pages = "686--690",
journal = "IEEE Geoscience and Remote Sensing Letters",
issn = "1545-598X",
publisher = "IEEE",
number = "5",

}

Weighted sparse graph based dimensionality reduction for hyperspectral images. / He, Wei; Zhang, Hongyan; Zhang, Liangpei; Philips, Wilfried; Liao, Wenzhi; Frery, Alejandro (Editor).

In: IEEE Geoscience and Remote Sensing Letters, Vol. 13, No. 5, 18.03.2016, p. 686-690.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Weighted sparse graph based dimensionality reduction for hyperspectral images

AU - He, Wei

AU - Zhang, Hongyan

AU - Zhang, Liangpei

AU - Philips, Wilfried

AU - Liao, Wenzhi

A2 - Frery, Alejandro

PY - 2016/3/18

Y1 - 2016/3/18

N2 - Dimensionality reduction (DR) is an important and helpful preprocessing step for hyperspectral image (HSI) classification. Recently, sparse graph embedding (SGE) has been widely used in the DR of HSIs. In this letter, we propose a weighted sparse graph based DR (WSGDR) method for HSIs. Instead of only exploring the locality structure (as in neighborhood preserving embedding) or the linearity structure (as in SGE) of the HSI data, the proposed method couples the locality and linearity properties of HSI data together in a unified framework for the DR of HSIs. The proposed method was tested on two widely used HSI data sets, and the results suggest that the locality and linearity are complementary properties for HSIs. In addition, the experimental results also confirm the superiority of the proposed WSGDR method over the other state-of-the-art DR methods.

AB - Dimensionality reduction (DR) is an important and helpful preprocessing step for hyperspectral image (HSI) classification. Recently, sparse graph embedding (SGE) has been widely used in the DR of HSIs. In this letter, we propose a weighted sparse graph based DR (WSGDR) method for HSIs. Instead of only exploring the locality structure (as in neighborhood preserving embedding) or the linearity structure (as in SGE) of the HSI data, the proposed method couples the locality and linearity properties of HSI data together in a unified framework for the DR of HSIs. The proposed method was tested on two widely used HSI data sets, and the results suggest that the locality and linearity are complementary properties for HSIs. In addition, the experimental results also confirm the superiority of the proposed WSGDR method over the other state-of-the-art DR methods.

KW - sparse graph embedding

KW - dimensionality reduction

KW - nearest neighbor graph

KW - hyperspectral image

KW - weighted sparse coding

KW - discriminant analysis

KW - feature extraction

U2 - 10.1109/LGRS.2016.2536658

DO - 10.1109/LGRS.2016.2536658

M3 - Article

VL - 13

SP - 686

EP - 690

JO - IEEE Geoscience and Remote Sensing Letters

T2 - IEEE Geoscience and Remote Sensing Letters

JF - IEEE Geoscience and Remote Sensing Letters

SN - 1545-598X

IS - 5

ER -