Classification of Hyperspectral Images by Gabor Filtering Based Deep Network

In this paper, a novel spectral-spatial classification method based on Gabor filtering and deep network (GFDN) is proposed. First, Gabor features are extracted by performing Gabor filtering on the first three principal components of the hyperspectral image, which can typically characterize the low-l...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE journal of selected topics in applied earth observations and remote sensing Jg. 11; H. 4; S. 1166 - 1178
Hauptverfasser: Kang, Xudong, Li, Chengchao, Li, Shutao, Lin, Hui
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Piscataway IEEE 01.04.2018
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
Classification
Deep learning
Feature extraction
Gabor filter
hyperspectral image (HSI) classification
Hyperspectral imaging
Image classification
Image filters
Image reconstruction
Machine learning
Methods
Spatial distribution
stacked sparse autoencoders (SSAE)
Training
virtual samples
ISSN:1939-1404, 2151-1535
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In this paper, a novel spectral-spatial classification method based on Gabor filtering and deep network (GFDN) is proposed. First, Gabor features are extracted by performing Gabor filtering on the first three principal components of the hyperspectral image, which can typically characterize the low-level spatial structures of different orientations and scales. Then, the Gabor features and spectral features are simply stacked to form the fused features. Afterwards, deep features are captured by training a stacked sparse autoencoder deep network with the fused features obtained above as inputs. Since the number of training samples of hyperspectral images is often very limited, which negatively affects the classification performance in deep learning, an effective way of constructing virtual samples is designed to generate more training samples, automatically. By jointly utilizing both the real and virtual samples, the parameters of the deep network can be better trained and updated, which can result in classification results of higher accuracies. Experiments performed on four real hyperspectral datasets show that the proposed method outperforms several recently proposed classification methods in terms of classification accuracies.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1939-1404
2151-1535
DOI:10.1109/JSTARS.2017.2767185