Unsupervised unmixing of hyperspectral imagery using the constrained positive matrix factorization

Abstract

In hyperspectral imaging, hundreds of images are taken at narrow and contiguous spectral bands providing us with high spectral resolution spectral signatures that can be used to discriminate between objects. In many applications, the measured spectral signature is a mixture of the object of interest, and other objects within the field of view of the sensor. To determine which objects are in the field of view of the sensor, we need to decompose the measured spectral signature in its constituents and their contribution to the measured signal. This research dealt with the unsupervised determination of the constituents and their fractional abundance in each pixel in a hyperspectral image using a constrained positive matrix factorization (cPMF). Different algorithms are presented to compute the cPMF. Tests and validation with real and simulated data show the effectiveness of the method. Application of the approach to environmental remote sensing and microscopic imaging is shown.