Assessing the Influence of Reference Spectra on Synthetic SAM Classification Results

Spectral matching algorithms, such as the Spectral Angle Mapper (SAM), utilize the spectral similarity between individual image pixel spectra and a spectral reference library with known components. Here, we illustrate and quantify the effects that different sources of reference libraries have on SAM...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on geoscience and remote sensing 2008-12, Vol.46 (12), p.4162-4172
Main Authors: Hecker, C., van der Meijde, M., van der Werff, H., van der Meer, F.D.
Format: Article
Language:English
Subjects:
Absorption
Acoustic microscopes
Applied geophysics
Classification
Earth sciences
Earth, ocean, space
Exact sciences and technology
Geology
Grounds
Hyperspectral imaging
Image analysis
Image classification
Infrared imaging
Infrared spectra
infrared spectroscopy
Internal geophysics
Libraries
Minerals
Pixel
Spectra
spectral analysis
Spectral Angle Mapper (SAM)
spectral library
spectral similarity measure
Studies
Testing
Wavelengths
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:Spectral matching algorithms, such as the Spectral Angle Mapper (SAM), utilize the spectral similarity between individual image pixel spectra and a spectral reference library with known components. Here, we illustrate and quantify the effects that different sources of reference libraries have on SAM classification results. Synthetic images of three mineral endmembers were classified by using reference libraries derived from airborne hyperspectral imagery, ground spectra (Portable Infrared Mineral Analyser), and from a standard library (United States Geologic Survey). Results show that the source of the reference library strongly influences the classification results if all available wavelengths are used. This effect can be partially neutralized by using appropriate preprocessing methods. Two different types of spectral subsetting of the data, two types of continuum removal, and a combination thereof were tested. Best results were achieved by using a feature subset (i.e., limiting the input wavelengths to the diagnostic absorption features). This increased the average classification accuracy from 74% to 95% (ground spectral library) and from 68% to 94% (standard library).
ISSN:0196-2892
1558-0644
DOI:10.1109/TGRS.2008.2001035