Parts | Days | Selection | Search | Updates | Downloads | Help

KFM: Fachverband Kristalline Festkörper und deren Mikrostruktur

KFM 6: Focus: Advanced TEM spectroscopy - low energy excitations and chemical composition at high resolution (joint session KFM/HL)

KFM 6.3: Talk

Monday, April 1, 2019, 15:50–16:10, PHY 5.0.20

Automatic Truncation of Principal Components in the PCA Analysis of EELS and EDX Spectrum-Images — •Pavel Potapov¹, Paolo Longo², and Axel Lubk¹ — ¹Leibniz Institute for Solid State and Materials Research (IFW), Dresden, Germany — ²Gatan Inc, Pleasanton, CA, USA

The Principal Component Analysis (PCA) allows to denoise drastically STEM EELS and EDX spectrum-images by extracting the meaningful fraction of data while cutting off the irrelevant noise. The number of meaningful PCA components is usually estimated through the evaluation of a scree plot - a dependence of the log eigenvalues (variances) on the component index. This strategy however introduces some subjectivity in the treatment. A novel promising method for the truncation of principal components is the analysis of bivariate scatter plots. This method can be easily implemented in automatic algorithms promoting a smooth, unsupervised data treatment flow.