Home / Research /

Spectral mixture analysis of EELS spectrum-images

When processing spectrum-images, microscopists usually try to map elemental, physical and chemical state information of a given material. This paper reports how a spectral unmixing (SU) algorithm dedicated to remote sensing hyperspectral images can be successfully applied to analyze spectrum-image resulting from electron energy-loss spectroscopy (EELS). SU generally overcomes standard limitations inherent to other multivariate statistical analysis methods, such as principal component analysis (PCA) or independent component analysis (ICA), that have been previously used to analyze EELS maps. Indeed, ICA and PCA may perform poorly for linear spectral mixture analysis due to the strong dependence between the abundances of the different materials. One example is presented here to demonstrate the potential of this technique for EELS analysis.

The Bayesian linear unmixing (BLU) algorithm and the main results are detailed in the paper published in Ultramicroscopy in 2012.

The Matlab codes of the BLU algorithm used in this study is available below.

New! A beta version of a Matlab friendly-interface (GUI) for the BLU algorithm:

Results on a real EELS spectrum-image

The analyzed dataset consists of a 64x64 pixel spectrum-image acquired in 1340 energy channels over a region composed of several nanocages in a boron-nitride nanotubes (BNNT) sample. Note that nanocages are supported by a holey carbon film for TEM analysis. These data have been extensively analyzed in a previous paper and a high angle dark field image of the region of interest is depicted in Fig. 1.
high angle dark field image
Fig. 1. HADF image corresponding to a 64x64 spectrum-image recorded in an area rich in nanoparticles containing boron (pure boron, boron oxide or h-BN).

The proposed algorithm has been applied on this real data set. The extracted spectral signatures are depicted in Fig. 2 and the corresponding abundance maps of the materials are depicted in Fig. 3.
Spectral components estimated by BLU
Fig. 2. Spectral components estimated by BLU. The recovered endmembers properly correspond to EELS spectra.
Maps of the spectral components
Fig. 3. Maps of the spectral components.

sitemeter stats