Comparison of Spectral Dissimilarity Measures and Dimension Reduction Techniques for Hyperspectral Images

Although hyperspectral data are becoming increasingly popular, they are difficult to use effectively due to the significant redundancy of such data. This article discusses a number of general-purpose dimensionality reduction techniques as a counter-redundancy measure that can be used in conjunction...

Full description

Saved in:
Bibliographic Details
Published in:Pattern recognition and image analysis 2021-07, Vol.31 (3), p.454-465
Main Author: Myasnikov, E. V.
Format: Article
Language:English
Subjects:
Analysis
Computer Science
Divergence
Euclidean geometry
Hyperspectral imaging
Image Processing and Computer Vision
Mathematical Theory of Images and Signals Representing
Pattern Recognition
Processing
Quality assessment
Recognition
Reduction
Redundancy
Spectra
Support vector machines
Understanding
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Although hyperspectral data are becoming increasingly popular, they are difficult to use effectively due to the significant redundancy of such data. This article discusses a number of general-purpose dimensionality reduction techniques as a counter-redundancy measure that can be used in conjunction with known spectral dissimilarity measures. The Euclidean distance, spectral angle, and divergence of spectral information are used as such dissimilarity measures. In order to map into a space of reduced dimension, we use nonlinear mapping (NLM), isomap, locally linear embedding (LLE), Laplacian eigenmaps, and uniform manifold approximation and projection (UMAP). Quality assessment is performed using well-known hyperspectral scenes based on the results obtained using the nearest neighbor (NN) classifier and support vector machine.
ISSN:1054-6618
1555-6212
DOI:10.1134/S1054661821030196