A hyperspectral unmixing framework for energy-loss near-edge structure analysis

Extracting different spectral components and their corresponding concentrations from spectrum images is one of the key challenges for electron energy-loss spectroscopy analysis due to the large amount of data, differing spectral features and low signal-to-noise ratio. Here, an open-source software framework of hyperspectral unmixing for energy-loss near-edge fine structure analysis is proposed. This software determines the number of independent spectral components, the signature of each spectral component and the abundance of each spectral component in each pixel, without reference spectrum or prior knowledge of the datasets. This approach should be suitable for automated materials and chemical analysis.