Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging

Jaime Zabalza, Jinchang Ren, Jiangbin Zheng, Huimin Zhao, Chunmei Qing, Zhijing Yang, Peijun Du, Stephen Marshall

Research output: Contribution to journalArticle

96 Citations (Scopus)

Abstract

Stacked autoencoders (SAEs), as part of the deep learning (DL) framework, have been recently proposed for feature extraction in hyperspectral remote sensing. With the help of hidden nodes in deep layers, a high-level abstraction is achieved for data reduction whilst maintaining the key information of the data. As hidden nodes in SAEs have to deal simultaneously with hundreds of features from hypercubes as inputs, this increases the complexity of the process and leads to limited abstraction and performance. As such, segmented SAE (S-SAE) is proposed by confronting the original features into smaller data segments, which are separately processed by different smaller SAEs. This has resulted in reduced complexity but improved efficacy of data abstraction and accuracy of data classification.

LanguageEnglish
Pages1-10
Number of pages10
JournalNeurocomputing
Volume185
Early online date23 Dec 2015
DOIs
Publication statusPublished - 12 Apr 2016

Fingerprint

Feature extraction
Remote sensing
Data reduction
Learning
Hyperspectral imaging
Deep learning
Data Accuracy

Keywords

  • data reduction
  • deep learning (DL)
  • hyperspectral remote sensing
  • segmented stacked autoencoder (S-SAE)

Cite this

Zabalza, Jaime ; Ren, Jinchang ; Zheng, Jiangbin ; Zhao, Huimin ; Qing, Chunmei ; Yang, Zhijing ; Du, Peijun ; Marshall, Stephen. / Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging. In: Neurocomputing. 2016 ; Vol. 185. pp. 1-10.
@article{60c7fd49c44545b096cf34efd0f576cb,
title = "Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging",
abstract = "Stacked autoencoders (SAEs), as part of the deep learning (DL) framework, have been recently proposed for feature extraction in hyperspectral remote sensing. With the help of hidden nodes in deep layers, a high-level abstraction is achieved for data reduction whilst maintaining the key information of the data. As hidden nodes in SAEs have to deal simultaneously with hundreds of features from hypercubes as inputs, this increases the complexity of the process and leads to limited abstraction and performance. As such, segmented SAE (S-SAE) is proposed by confronting the original features into smaller data segments, which are separately processed by different smaller SAEs. This has resulted in reduced complexity but improved efficacy of data abstraction and accuracy of data classification.",
keywords = "data reduction, deep learning (DL), hyperspectral remote sensing, segmented stacked autoencoder (S-SAE)",
author = "Jaime Zabalza and Jinchang Ren and Jiangbin Zheng and Huimin Zhao and Chunmei Qing and Zhijing Yang and Peijun Du and Stephen Marshall",
year = "2016",
month = "4",
day = "12",
doi = "10.1016/j.neucom.2015.11.044",
language = "English",
volume = "185",
pages = "1--10",
journal = "Neurocomputing",
issn = "0925-2312",

}

Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging. / Zabalza, Jaime; Ren, Jinchang; Zheng, Jiangbin; Zhao, Huimin; Qing, Chunmei; Yang, Zhijing; Du, Peijun; Marshall, Stephen.

In: Neurocomputing, Vol. 185, 12.04.2016, p. 1-10.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging

AU - Zabalza, Jaime

AU - Ren, Jinchang

AU - Zheng, Jiangbin

AU - Zhao, Huimin

AU - Qing, Chunmei

AU - Yang, Zhijing

AU - Du, Peijun

AU - Marshall, Stephen

PY - 2016/4/12

Y1 - 2016/4/12

N2 - Stacked autoencoders (SAEs), as part of the deep learning (DL) framework, have been recently proposed for feature extraction in hyperspectral remote sensing. With the help of hidden nodes in deep layers, a high-level abstraction is achieved for data reduction whilst maintaining the key information of the data. As hidden nodes in SAEs have to deal simultaneously with hundreds of features from hypercubes as inputs, this increases the complexity of the process and leads to limited abstraction and performance. As such, segmented SAE (S-SAE) is proposed by confronting the original features into smaller data segments, which are separately processed by different smaller SAEs. This has resulted in reduced complexity but improved efficacy of data abstraction and accuracy of data classification.

AB - Stacked autoencoders (SAEs), as part of the deep learning (DL) framework, have been recently proposed for feature extraction in hyperspectral remote sensing. With the help of hidden nodes in deep layers, a high-level abstraction is achieved for data reduction whilst maintaining the key information of the data. As hidden nodes in SAEs have to deal simultaneously with hundreds of features from hypercubes as inputs, this increases the complexity of the process and leads to limited abstraction and performance. As such, segmented SAE (S-SAE) is proposed by confronting the original features into smaller data segments, which are separately processed by different smaller SAEs. This has resulted in reduced complexity but improved efficacy of data abstraction and accuracy of data classification.

KW - data reduction

KW - deep learning (DL)

KW - hyperspectral remote sensing

KW - segmented stacked autoencoder (S-SAE)

UR - http://www.scopus.com/inward/record.url?scp=84961344134&partnerID=8YFLogxK

UR - http://www.sciencedirect.com/science/journal/09252312

U2 - 10.1016/j.neucom.2015.11.044

DO - 10.1016/j.neucom.2015.11.044

M3 - Article

VL - 185

SP - 1

EP - 10

JO - Neurocomputing

T2 - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

ER -