Dresden 2020 – scientific programme

The DPG Spring Meeting in Dresden had to be cancelled! Read more ...

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

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 103: Topical Session: Data Driven Materials Science - Machine Learning for Materials Characterization (joint session MM/CPP)

CPP 103.2: Talk

Thursday, March 19, 2020, 16:15–16:30, BAR 205

Automatic semantic segmentation of Scanning Transmission Electron Microscopy (STEM) images using an unsupervised machine learning approach — •Ning Wang, Christoph Freysoldt, Christian Liebscher, and Jörg Neugebauer — Max-Planck-Institut für Eisenforschung GmbH, Düsseldorf

The recent substantial advance of machine learning provides us with a rich toolbox to successfully address problems in materials science. Here we present an unsupervised machine learning approach for automatic semantic segmentation of STEM images. We propose a robust descriptor, the local correlation map, for characterization of the local periodicity, which is then fed into an unsupervised clustering algorithm in order to segment the STEM images into different crystalline regions. The semantic segmentation works as an initial step for further data analysis, such as image denoising, extraction of lattice vectors and so on. As a proof of concept, we apply our approach to STEM images of Cu grain boundaries, Ni stacking faults and twin boundaries, and Fe2Nb phase boundaries, and observe very good robustness and resolution.