Spectral unmixing of three-algae mixtures using hyperspectral images

Abstract

A hyperspectral imaging system has been used to acquire hyperspectral data representing various combinations of three pure algal mixtures in liquid media. Geometric and linear spectral unmixing methods have been applied to identify the ratiometric combinations of the algae in the mixtures. For the geometric method, two local spectral slopes have been identified as spectral features. Average feature values for each class of algae are used as vertices of a triangle, and then compared to the test features to predict algal ratios in the test mixture. The results are compared to those from classic linear spectral unmixing. In the two independent data sets prepared, the introduced geometric method produced more favorable results than the classical spectral unmixing method.