Deep Filter Banks for Land-Use Scene Classification

Land-use (LU) scene classification is one of the most challenging tasks in the field of remote sensing (RS) image processing due to its high intraclass variability and low interclass distance. Motivated by the challenge posed by this problem, we propose a novel hybrid architecture, deep filter banks...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE geoscience and remote sensing letters Jg. 13; H. 12; S. 1895 - 1899
Hauptverfasser: Wu, Hang, Liu, Baozhen, Su, Weihua, Zhang, Wenchang, Sun, Jinggong
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Piscataway IEEE 01.12.2016
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
Classification
Data acquisition
Data models
Deep filter banks
Encoding
Feature extraction
Filter banks
Filters
Fisher vector (FV)
Image processing
Kernel
Land use
land-use (LU) scene classification
Learning algorithms
Machine learning
Noise reduction
Remote sensing
Robustness
Semantics
stacked denoising sparse autoencoder (SDSAE)
ISSN:1545-598X, 1558-0571
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Land-use (LU) scene classification is one of the most challenging tasks in the field of remote sensing (RS) image processing due to its high intraclass variability and low interclass distance. Motivated by the challenge posed by this problem, we propose a novel hybrid architecture, deep filter banks, combining multicolumn stacked denoising sparse autoencoder (SDSAE) and Fisher vector (FV) to automatically learn the representative and discriminative features in a hierarchical manner for LU scene classification. SDSAE kernels describe local patches and a robust global feature of the RS image is built through the FV pooling layer. Unlike previous handcrafted features, we use machine-learning mechanisms to optimize our proposed feature extractor so that it can learn more suitable internal features from the RS data, boosting the final performance. Our approach achieves superior performance compared with the state-of-the-art methods, obtaining average classification accuracies of 92.7% and 90.4%, respectively, on the UC Merced and RSSCN7 data sets.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1545-598X
1558-0571
DOI:10.1109/LGRS.2016.2616440