mcrals: Angle constrained MCR-ALS.

In svkucheryavski/supure: Spectral Unmixing with Pure Variables

Description

Angle constrained MCR-ALS.

Usage

mcrals(D, ncomp, Ct = NULL, St = NULL, do.angle = T, f = 0.05,
  wavelength = NULL, max.niter = 500)

Arguments

D	a matrix with spectral values
ncomp	number of pure components to identify
Ct	initial estimates for pure contributions
St	initial estimates for pure spectra
do.angle	logical, should the angle constraint be used
f	how much of mean should be added for angle constraint
wavelength	a vector with wavelength if necessary
max.niter	maximum number of iterations

Details

The method implements a simple MCR-ALS algorithm with non-negativity constraints applied both to spectra and contributions by setting all negative values to zero. In addition an angle constraint can be applied by adding a small portion (default 5

If initial estimates for contributions and spectra are not provided the estimates will be generated with random numbers.

Value

The method returns results of unmixing as an object of mcrals class with following fields.

spec	resolved spectra of pure components.
conc	resolved concentrations of pure components.
ncomp	number of resolved components.
resvar	residual variance for each component.

References

1. W. Windig, M.R. Keenan, Appl. Spectrosc. 65 (2011) 349-357.

See Also

The class has numerous methods for investigating and refininf the unmixing results.

print	prints information about a purity object.
summary	shows summary statistics for the object.
plot	plot overview of unmixing results.
plotSpectra	shows a plot with resolved spectra.
plotConcentrations	shows a plot with resolved concentrations.
plotVariance	shows a plot with residual variance vs. number of components.
plotSpectraComparison	compares resolved and ethalon spectra if available.
image.mcrals	if data is a hypercube, shows resolved concetration map as an image.

Examples

# unmix hyperspectral data with and without angle constraint and compare the
# results for first two components
data(puredata)

res1 = mcrals(hsi$D, 3, do.angle = FALSE, wavelength = hsi$wavelength)
res2 = mcrals(hsi$D, 3, do.angle = TRUE, wavelength = hsi$wavelength)

summary(res1)
summary(res2)

# because initial estimates are made with random numbers
# the order of components in the results is different
par(mfrow = c(2, 1))
plotSpectra(res1)
plotSpectra(res2)
par(mfrow = c(1, 1))

par(mfrow = c(3, 2))
for (i in 1:3)
{
 image(res1, i)
 image(res2, i)
}
par(mfrow = c(1, 1))

svkucheryavski/supure documentation built on May 30, 2019, 9:32 p.m.