MIMR-DGSA

unsupervised hyperspectral band selection based on information theory and a modified discrete gravitational search algorithm

Julius Tschannerl, Jinchang Ren, Peter Yuen, Genyun Sun, Huimin Zhao, Zhijing Yang, Zheng Wang, Stephen Marshall

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

Band selection plays an important role in hyperspectral data analysis as it can improve the performance of data analysis without losing information about the constitution of the underlying data. We propose a MIMR-DGSA algorithm for band selection by following the Maximum-Information-Minimum-Redundancy (MIMR) criterion that maximises the information carried by individual features of a subset and minimises redundant information between them. Subsets are generated with a modified Discrete Gravitational Search Algorithm (DGSA) where we definine a neighbourhood concept for feature subsets. A fast algorithm for pairwise mutual information calculation that incorporates variable bandwidths of hyperspectral bands called VarBWFastMI is also developed. Classification results on three hyperspectral remote sensing datasets show that the proposed MIMR-DGSA performs similar to the original MIMR with Clonal Selection Algorithm (CSA) but is computationally more efficient and easier to handle as it has fewer parameters for tuning.
Original languageEnglish
Pages (from-to)189-200
Number of pages12
JournalInformation Fusion
Volume51
Early online date15 Feb 2019
DOIs
Publication statusPublished - 1 Nov 2019

Fingerprint

Information theory
Redundancy
Set theory
Remote sensing
Tuning
Bandwidth

Keywords

  • band selection
  • discrete optimisation
  • entropy
  • evolutionary computation
  • feature selection
  • gravitational search algorithm
  • hyperspectral imaging
  • maximum-information-minimum-redundancy
  • mutual information.

Cite this

@article{1c066443a1da49e5b9c04e025de08445,
title = "MIMR-DGSA: unsupervised hyperspectral band selection based on information theory and a modified discrete gravitational search algorithm",
abstract = "Band selection plays an important role in hyperspectral data analysis as it can improve the performance of data analysis without losing information about the constitution of the underlying data. We propose a MIMR-DGSA algorithm for band selection by following the Maximum-Information-Minimum-Redundancy (MIMR) criterion that maximises the information carried by individual features of a subset and minimises redundant information between them. Subsets are generated with a modified Discrete Gravitational Search Algorithm (DGSA) where we definine a neighbourhood concept for feature subsets. A fast algorithm for pairwise mutual information calculation that incorporates variable bandwidths of hyperspectral bands called VarBWFastMI is also developed. Classification results on three hyperspectral remote sensing datasets show that the proposed MIMR-DGSA performs similar to the original MIMR with Clonal Selection Algorithm (CSA) but is computationally more efficient and easier to handle as it has fewer parameters for tuning.",
keywords = "band selection, discrete optimisation, entropy, evolutionary computation, feature selection, gravitational search algorithm, hyperspectral imaging, maximum-information-minimum-redundancy, mutual information.",
author = "Julius Tschannerl and Jinchang Ren and Peter Yuen and Genyun Sun and Huimin Zhao and Zhijing Yang and Zheng Wang and Stephen Marshall",
year = "2019",
month = "11",
day = "1",
doi = "10.1016/j.inffus.2019.02.005",
language = "English",
volume = "51",
pages = "189--200",
journal = "Information Fusion",
issn = "1566-2535",

}

MIMR-DGSA : unsupervised hyperspectral band selection based on information theory and a modified discrete gravitational search algorithm. / Tschannerl, Julius; Ren, Jinchang; Yuen, Peter; Sun, Genyun; Zhao, Huimin; Yang, Zhijing; Wang, Zheng; Marshall, Stephen.

In: Information Fusion , Vol. 51, 01.11.2019, p. 189-200.

Research output: Contribution to journalArticle

TY - JOUR

T1 - MIMR-DGSA

T2 - unsupervised hyperspectral band selection based on information theory and a modified discrete gravitational search algorithm

AU - Tschannerl, Julius

AU - Ren, Jinchang

AU - Yuen, Peter

AU - Sun, Genyun

AU - Zhao, Huimin

AU - Yang, Zhijing

AU - Wang, Zheng

AU - Marshall, Stephen

PY - 2019/11/1

Y1 - 2019/11/1

N2 - Band selection plays an important role in hyperspectral data analysis as it can improve the performance of data analysis without losing information about the constitution of the underlying data. We propose a MIMR-DGSA algorithm for band selection by following the Maximum-Information-Minimum-Redundancy (MIMR) criterion that maximises the information carried by individual features of a subset and minimises redundant information between them. Subsets are generated with a modified Discrete Gravitational Search Algorithm (DGSA) where we definine a neighbourhood concept for feature subsets. A fast algorithm for pairwise mutual information calculation that incorporates variable bandwidths of hyperspectral bands called VarBWFastMI is also developed. Classification results on three hyperspectral remote sensing datasets show that the proposed MIMR-DGSA performs similar to the original MIMR with Clonal Selection Algorithm (CSA) but is computationally more efficient and easier to handle as it has fewer parameters for tuning.

AB - Band selection plays an important role in hyperspectral data analysis as it can improve the performance of data analysis without losing information about the constitution of the underlying data. We propose a MIMR-DGSA algorithm for band selection by following the Maximum-Information-Minimum-Redundancy (MIMR) criterion that maximises the information carried by individual features of a subset and minimises redundant information between them. Subsets are generated with a modified Discrete Gravitational Search Algorithm (DGSA) where we definine a neighbourhood concept for feature subsets. A fast algorithm for pairwise mutual information calculation that incorporates variable bandwidths of hyperspectral bands called VarBWFastMI is also developed. Classification results on three hyperspectral remote sensing datasets show that the proposed MIMR-DGSA performs similar to the original MIMR with Clonal Selection Algorithm (CSA) but is computationally more efficient and easier to handle as it has fewer parameters for tuning.

KW - band selection

KW - discrete optimisation

KW - entropy

KW - evolutionary computation

KW - feature selection

KW - gravitational search algorithm

KW - hyperspectral imaging

KW - maximum-information-minimum-redundancy

KW - mutual information.

UR - https://www.sciencedirect.com/journal/information-fusion

U2 - 10.1016/j.inffus.2019.02.005

DO - 10.1016/j.inffus.2019.02.005

M3 - Article

VL - 51

SP - 189

EP - 200

JO - Information Fusion

JF - Information Fusion

SN - 1566-2535

ER -