| US 7,567,712 B2 | ||
| Method of identifying endmember spectral values from hyperspectral image data | ||
| Mark Berman, Pymble (Australia); Haari Taplo Kiiveri, Bullcreek (Australia); Ryan Lagerstrom, Glebe (Australia); and Robert Aldan Dunne, Darlington (Australia) | ||
| Assigned to Commonwealth Scientific and Industrial Research Organisation, Campbell Act (Australia) | ||
| Appl. No. 10/510,078 PCT Filed Apr. 04, 2003, PCT No. PCT/AU03/00409 § 371(c)(1), (2), (4) Date Dec. 14, 2004, PCT Pub. No. WO03/088130, PCT Pub. Date Oct. 23, 2003. |
||
| Claims priority of application No. PS1552 (AU), filed on Apr. 05, 2002. | ||
| Prior Publication US 2005/0180634 A1, Aug. 18, 2005 | ||
| Int. Cl. G06K 9/46 (2006.01) | ||
| U.S. Cl. 382—191 | 13 Claims |
| 1. A method of identifying endmember spectra values from a multispectral image comprising multispectral image data, where
each multispectral data value is equal to a sum of mixing proportions of each endmember spectrum, said method including the
steps of:
processing the multispectral image data to obtain a multidimensional simplex having a number of vertices equal to the number
of endmembers, the position of each vertex representing a spectrum of one of the endmembers, wherein processing of the data
includes:
providing starting estimates of each endmember spectrum for each image data value;
estimating mixing proportions for each data value from estimates of the spectra of all the endmembers;
estimating the spectrum of each endmember from the estimates of the mixing proportions of the spectra of all the endmembers
for each image data value; and
repeating the estimation of the mixing proportions and the estimation of the spectrum of each endmember until a stopping condition
is met, wherein the stopping condition occurs when a relative change in a regularized residual sum of squares determined in
the estimation steps attains a threshold,
wherein the regularized residual sum of squares comprises a sum of residual sum of squares and a measure of the size of the
simplex, the residual squares is reflective of a difference between the multispectral image data and a calculated value based
on the estimated mixing proportions and estimated spectrum of each endmember.
|