Non-negative factor analysis supporting the interpretation of elemental distribution images acquired by XRF

22nd International Congress on X-Ray Optics and Microanalysis

Alfeld, Matthias^1 ; Wahabzada, Mirwaes^2 ; Bauckhage, Christian^2 ; Kersting, Kristian^2,3 ; Wellenreuther, Gerd^4 ; Falkenberg, Gerald^1

Photon Science, DESY, Notkestraße 85, 22607 Hamburg, Germany^1

Fraunhofer IAIS, Schloss Birlinghoven, 53757 Sankt Augustin, Germany^2

Computer Science Department, TU Dortmund University, Joseph-von-Fraunhofer-Str. 23, 44227 Dortmund, Germany^3

European XFEL GmbH, Albert-Einstein-Ring 19, 22761 Hamburg, Germany^4

关键词: Elemental compositions;    Elemental distribution;    Matrix factorizations;    Negative values;    Non-negativity;    Nonnegative matrix factorization;    Priori knowledge;    User intervention;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/499/1/012013/pdf
DOI  :  10.1088/1742-6596/499/1/012013

来源: IOP

【 摘 要 】

Stacks of elemental distribution images acquired by XRF can be difficult to interpret, if they contain high degrees of redundancy and components differing in their quantitative but not qualitative elemental composition. Factor analysis, mainly in the form of Principal Component Analysis (PCA), has been used to reduce the level of redundancy and highlight correlations. PCA, however, does not yield physically meaningful representations as they often contain negative values. This limitation can be overcome, by employing factor analysis that is restricted to non-negativity. In this paper we present the first application of the Python Matrix Factorization Module (pymf) on XRF data. This is done in a case study on the painting Saul and David from the studio of Rembrandt van Rijn. We show how the discrimination between two different Co containing compounds with minimum user intervention and a priori knowledge is supported by Non-Negative Matrix Factorization (NMF).