Understanding multi-spectral images of woodparticles with matrix factorization

Multispectral image data can be used to quantify the
concentrations of chemical substances in material compounds
by differential spectroscopy. In this paper, we describe Simplex
Volume Maximization (SiVM), a matrix factorization method derived
from Archetypal Analysis (AA), that is well suited to separate
spectra. Exemplarily, we apply the technique to multispectral
images of wood strands partially covered with adhesives and
wood-polymer composites and show how to determine the concentration
of the adhesives and how to distinguish the polymer
types.
In the multispectral domain, our objective is to separate the spectral
characteristics of the adhesives and polymers from those
spectral components caused by variation in the natural wood, including
differences in moisture.
Our experiments show that wood particles with different concentrations
of adhesives or different polymer components can be distinguished
after applying SiVM-based factorization to NIR spectral
imaging. We therefore conclude that this technique has great
potential for quality control applications that rely on multispectral
imaging.