A4 Conference proceedings

A multi-manifold clustering algorithm for hyperspectral remote sensing imagery

Open Access publication

LUT Authors / Editors

Publication Details
Authors: Hassanzadeh Aidin, Kauranne Tuomo, Kaarna Arto
Publication year: 2016
Language: English
Related Journal or Series Information: IEEE International Geoscience and Remote Sensing Symposium proceedings
Title of parent publication: Proceedings of Geoscience and Remote Sensing (IGARSS), IEEE International Symposium
Start page: 3326
End page: 3329
ISBN: 978-1-5090-3333-1
eISBN: 978-1-5090-3332-4
ISSN: 2153-7003
JUFO-Level of this publication: 1
Open Access: Open Access publication

Unsupervised classification plays a key role in remote sensing hyperspectral image analysis. Complexities arise from the high dimensionality of hyperspectral imagery and this implies the need for dimensionality reduction as a vital preprocessing step. However, conventional dimensionality reduction techniques, such as linear and nonlinear manifold learning approaches, may fail if the hyperspectral remote sensing data stem from several intersecting data manifolds. In this paper, we consider remote sensing hyperspectral data within the framework of multi-manifold learning with possible intersections. To this end, we propose a multi-manifold spectral clustering algorithm for unsupervised classification of hyperspectral imagery. The proposed algorithm exploits the notion of shared nearest neighbourhood for the construction of nearest neighbour connectivity model and a weighted principal component analysis model for tangent space estimation. Preliminary results on two benchmark hyperspectral data sets reveal the superiority of the proposed algorithm in terms of clustering accuracy over approaches based on conventional dimensionality reduction techniques.

Last updated on 2018-19-10 at 08:49