Publication Cover
Canadian Journal of Remote Sensing
Journal canadien de télédétection
Volume 44, 2018 - Issue 5
110
Views
3
CrossRef citations to date
0
Altmetric
Original Articles

Hyperspectral Image Classification Based on Fusion of Guided Filter and Domain Transform Interpolated Convolution Filter

, , &
Pages 476-490 | Received 05 Sep 2018, Accepted 06 Nov 2018, Published online: 08 Mar 2019
 

Abstract

In recent years, the spatial texture features obtained by filtering have become a hot research topic to improve hyperspectral image classification, but spatial correlation information is often lost in spatial texture information extraction. To solve this problem, a spectral-spatial classification method based on guided filtering and by the algorithm Large Margin Distribution Machine (LDM) is proposed. More specifically, the spatial texture features can be extracted by a Guided filter (GDF) from hyperspectral images whose dimensionality is reduced by a Principal Component Analysis (PCA). Spatial correlation features of the hyperspectral image are then obtained using a Domain Transform Interpolated Convolution Filter. The last step is to fuse spatial texture features and correlation features for classification by LDM. The experimental results using the actual hyperspectral image indicate that the proposed GDFDTICF-LDM method is superior to other classification methods, such as the original Support Vector Machine (SVM) with raw spectral features, dimensionality reduction features and spatial-spectral information, methods of edge-preserving filter and recursive filter, and LDM-based methods.

RÉSUMÉ

Ces dernières années, les caractéristiques de la texture spatiale obtenues par filtrage sont devenues un sujet de recherche populaire et ce afin d’améliorer la classification d’images hyperspectrales. Toutefois, les corrélations spatiales sont souvent perdues dans l’extraction d’information sur la texture spatiale. Pour résoudre ce problème, on propose une méthode de classification spectrale-spatiale basée sur un filtrage guidée et l’algorithme (Large Margin Distribution Machine (LDM)). Plus précisément, les caractéristiques spatiales de la texture peuvent être extraites par un filtre (GDF) d’images hyperspectrales dont la dimensionnalité est réduite par l’ACP. Les caractéristiques de la corrélation spatiale de l’image hyperspectrale sont alors obtenues en utilisant un filtre appelé Domain Transform Interpolated Convolution (DTICF). La dernière étape consiste à fusionner les caractéristiques spatiales de texture et de corrélation pour la classification LDM. Les résultats expérimentaux sur une image hyperspectrale réelle indiquent que la méthode GDFDTICF-LDM proposée est supérieure aux autres méthodes de classification, comme le SVM original avec des caractéristiques spectrales brutes, les fonctions de réduction de la dimensionnalité spatiale et spectrale, les filtres de préservation des contours, les filtres récursifs et les méthodes LDM.

Additional information

Funding

This work was supported by National Natural Science Foundation of China (Grant No 61275010, 61675051, 61701272 and 61701123), Natural Science Foundation of Guangdong (Grant No 2018A030313195), Major research project of Guangdong (Grant No 2017GKTSCX021). Science and Technology Project of Guangzhou (Grant No 201804010262), Science and Technology Project of Guangdong (Grant No 2017ZC0358), Fundamental Research Funds of Guangdong Communication Polytechnic (Grant No 2017-1-001).

Log in via your institution

Log in to Taylor & Francis Online

PDF download + Online access
  • 48 hours access to article PDF & online version
  • Article PDF can be downloaded
  • Article PDF can be printed
USD 61.00 Add to cart
* Local tax will be added as applicable

Related Research

People also read lists articles that other readers of this article have read.

Recommended articles lists articles that we recommend and is powered by our AI driven recommendation engine.

Cited by lists all citing articles based on Crossref citations.
Articles with the Crossref icon will open in a new tab.