Title :
Beyond the resolution limit: using least squares for subpixel analysis in remote sensing
Author_Institution :
Dept. of Comput. & Appl. Math., Univ. of the Witwatersrand, Johannesburg, South Africa
Abstract :
Spectral unmixing against a library of known endmembers can be modelled as a linear least squares problem with constraints. We take a different approach: model parameters are mapped through the log-odds transformation into a space where maximum likelihood parameter estimation leads to an unconstrained nonlinear least squares problem. Newton´s method is then proposed for its resolution
Keywords :
Newton method; image resolution; least squares approximations; maximum likelihood estimation; remote sensing; Newton´s method; linear least squares problem; log-odds transformation; maximum likelihood parameter estimation; model parameters; remote sensing; resolution limit; satellite image; spatial resolution; spectral unmixing; subpixel analysis; unconstrained nonlinear least squares problem; Geologic measurements; Geology; Least squares approximation; Least squares methods; Libraries; Predictive models; Remote sensing; Satellites; Spatial resolution; Vegetation mapping;
Conference_Titel :
AFRICON, 1996., IEEE AFRICON 4th
Conference_Location :
Stellenbosch
Print_ISBN :
0-7803-3019-6
DOI :
10.1109/AFRCON.1996.562956
