Global mean values in linear spectral unmixing: double fallacy!

Authors:
T. G. Ngigi;R. Tateishi;M. Gachari
Affiliations:
Center for Environmental Remote Sensing, Chiba University, Chiba, 263-8522, Japan;Center for Environmental Remote Sensing, Chiba University, Chiba, 263-8522, Japan;Kimathi University College, Jomo Kenyatta University of Agriculture and Technology, Kenya
Venue:
International Journal of Remote Sensing
Year:
2009

Citing 2
Cited 0

Comparison of a new classifier, the Mix-Unmix Classifier, with conventional hard and soft classifiers

International Journal of Remote Sensing
Modern computational techniques for environmental data; application to the global ozone layer

ICCS'05 Proceedings of the 5th international conference on Computational Science - Volume Part III

Quantified Score

Hi-index	0.00

Visualization

Abstract

Almost all conventional linear spectral unmixing techniques are based on the principle of least squares. The global mean digital number (DN) of an end-member is taken as the representative (i.e. contributory) DN for the end-member. This paper sets out to prove that the notion is a fallacy, and will always lead to negative percentages, super-positive percentages and non-100% sum of percentages if the unmixed pixel is not composed of, to within some tolerance, the global mean DNs only. Three sets of spectral end-members (two, three and four spectral end-members) are generated from Landsat ETM+ data. Practical percentages (between 0% and 100% and totalling 100%) of the end-members are returned by pixels in which the local mean DNs of the spectral end-members do not differ from the global mean DNs by, on average, 4.