SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 87: Focus Session: Scanning Probe Microscopy with Quartz Sensors III
O 87.4: Vortrag
Donnerstag, 30. März 2023, 16:00–16:15, TRE Ma
Machine learning for high-resolution AFM image interpretation — •Niko Oinonen1, Lauri Kurki1, Chen Xu1, Shuning Cai1, Markus Aapro1, Alexander Ilin2, Peter Liljeroth1, and Adam Foster1,3 — 1Department of Applied Physics, Aalto University, Finland — 2Department of Computer Science, Aalto University, Finland — 3WPI Nano Life Science Institute (WPI-NanoLSI), Kanazawa University, Japan
State-of-the-art non-contact atomic force microscopy (AFM) setups operating in vacuum at low temperatures are able to resolve features on the scale of individual atoms in molecules [1]. However, the process of interpreting the resulting AFM images is often a very challenging task even for highly trained experts in the field. We are working towards greater interpretability and greater automation of the processing of AFM images using machine learning methods. We have introduced an approach based on convolutional neural networks for discovering the atomic structure and electrostatic properties of samples directly from AFM images via image descriptors that characterize the sample [2]. Our recent work refines the geometry prediction task by predicting the molecule graph of the sample using a model based on graph neural networks [3]. The current challenge is in generalizing from simulated training data to experimental test data, where we find that the choice of the training data becomes very important.
[1] L. Gross et al., Science, vol. 325, no. 5944, pp. 1110–1114, 2009.
[2] N. Oinonen et al., ACS Nano, 16, 1, 89–97, 2022.
[3] N. Oinonen et al., MRS Bulletin 47, 895–905, 2022.