SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DS: Fachverband Dünne Schichten
DS 4: Thin Film Properties I
DS 4.4: Vortrag
Montag, 27. März 2023, 16:15–16:30, SCH A 316
Deep learning-supported in-situ HRTEM experiments on single-layer carbon — •Christopher Leist1, Haoyuan Qi1,2, and Ute Kaiser1 — 1Central Facility Materials Science Electron Microscopy, Ulm University, 89081 Ulm, Germany — 2Faculty of Chemistry and Food Chemistry Dresden, Technische Universität Dresden, 01062 Dresden, Germany.
We perform in-situ experiments on single-layer carbon using the Cc/Cs-corrected low-voltage transmission electron microscope SALVE, resolving structure and dynamics down to the level of the single atom. These types of experiments create large amounts of data both in terms of numbers of individual images acquired and the amount of information per image. Conventional image analysis methods, e.g., handcrafted filter kernels, often require heavy user supervision and tremendous time cost, posing strong limitations on the data volume making them inconvenient for use in these experiments. Deep learning in the form of convolutional neural networks offers a reliable and effective way to handle large amounts of complex image data. Using simulated data we train a modified U-Net like neural network to identify atom positions and their structure i.e. polygons while at the same time removing contaminated areas from the evaluation in real micrographs. Thus, gaining the ability for both largescale statistical evaluation and mapping atomically-resolved the material's transformation.