Performance analysis of unsupervised unmixing models for thermal hyepsrpectral

Hyperspectral unmixing which consists of decomposing the measured pixel radiance into mixtures of `pure´ spectra whose fractions are referred to as abundances, is a common procedure in the signal and image applications. In this paper, the performance of unsupervised unmixing methods for hyperspectral data is evaluated using airborne thermal data and pixel-based ground truth in-situ. Specifically, the procedures of endmember extraction and mixing using linear and bilinear models are validated using set of pixels whose mixing proportion are known.