NCAFM2023 Programme Booklet

AUTOMATIC IDENTIFICATION OF APPARENT ATOM POSITIONS IN BOND RESOLVED AFM IMAGES

Alexander Riss 1,*

1 Physics Department E20, Technical University of Munich, D-85748 Garching, Germany

Email: a.riss@tum.de

Apparent atom positions and apparent bond-lengths in bond-resolved atomic force microscopy (AFM) images of organic molecules can reveal information on the chemical and physical properties of the compounds [1-4]. Typically, such bond length measurements are done by hand, which is a very time-intensive task, thus strongly limiting the systematic analysis of large datasets. In this work, we present an approach for automatic identification of apparent atom positions in bond-resolved AFM images. As generating labeled datasets of sufficient quality and size to train neural networks can be challenging, we use an approach based on multiple steps of conventional image processing algorithms. The extracted positions are of sufficient quality for direct use for molecular structure recognition of organic compounds. Furthermore, we provide a simple and user-friendly graphical user interface to manually refine the apparent atom positions for the precise analysis of bond lengths.

Fig. 1 : Automatic identification of apparent atom positions in a bond-resolved AFM image of an organic molecule.

References [1] L. Gross et al. Science 337 , 1326 (2012) [2] P. Hapala et al. PRL 113 , 226101 (2014) [3] P. Hapala et al. Nat. Comm. 7 , 11560 (2016) [4] S. Fatayer et al. Science 365 , 6449 (2019)

101