WhatEELS. A python-based interactive software solution for ELNES analysis combining clustering and NLLS

• Published in: Ultramicroscopy 2022-01, Vol.232, p.113403-113403, Article 113403
• Main Authors: Blanco-Portals, J., Torruella, P., Baiutti, F., Anelli, S., Torrell, M., Tarancón, A., Peiró, F., Estradé, S.
• Format: Article
• Language: English
• Subjects: Clustering analysis; Electron energy loss spectroscopy (EELS); Elemental quantification; Energy loss near edge structure (ELNES); Non-linear least squares fitting (NLLS); Oxidation state analysis
• ISSN: 0304-3991; 1879-2723
• DOI: 10.1016/j.ultramic.2021.113403

•Combination of clustering an NLLS for energy loss near-edge spectroscopy analysis.•Free, interactive and modular software tool for spectral analysis based on Python.•Results analysis routines for oxidation state quantification from EELS datasets. The analysis of energy loss near edge structures in EELS is a powerful method for a precise characterization of elemental oxidation states and local atomic coordination with an outstanding lateral resolution, down to the atomic scale. Given the complexity and sizes of the EELS spectrum images datasets acquired by the state-of-the-art instrumentation, methods with low convergence times are usually preferred for spectral unmixing in quantitative analysis, such as multiple linear least squares fittings. Nevertheless, non-linear least squares fitting may be a superior choice for analysis in some cases, as it eliminates the need of calibrated reference spectra and provides information for each of the individual components included in the fitted model. To avoid some of the problems that the non-linear least squares algorithms may suffer dealing with mixed-composition samples and, thus, a model comprised by a large number of individual curves we proposed the combination of clustering analysis for segmentation and non-linear least squares fitting for spectral analysis. Clustering analysis is capable of a fast classification of pixels in smaller subsets divided by their spectral characteristics, and thus increases the control over the model parameters in separated regions of the samples, classified by their specific compositions. Furthermore, along with this manuscript we provide access to a self-contained and expandable modular software solution called WhatEELS. It was specifically designed to facilitate the combined use of clustering and NLLS, and includes a set of tools for white-lines analysis and elemental quantification. We successfully demonstrated its capabilities with a control sample of mesoporous cerium oxide doped with praseodymium and gadolinium, which posed challenging case-study given its spectral characteristics.