Regensburg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 17: Poster Monday: Organic Molecules at Surfaces 1
O 17.5: Poster
Montag, 5. September 2022, 18:00–20:00, P4
Generation of synthetic chiral structured images for computer vision applications — •Johannes Tim Seifert1, Peer Kasten1, Mandy Stritzke2, Björn Möller3, Timo de Wolff2, Tim Fingscheidt3, and Uta Schlickum1 — 1Institut für Angewandte Physik, Technische Universität Braunschweig — 2Institut für Analysis und Algebra, Technische Universität Braunschweig — 3Institut für Nachrichtentechnik, Technische Universität Braunschweig
Scanning tunneling microscopy (STM) is a well-established tool to measure surface topographs with atomic precision. Since data generation is slow using STM, we present a tool to create synthetic images with realistic noise and defects, which provides labeled training data for neural networks. We compare two methods evaluating chiral structures in STM images using Deep Learning.
As one approach to classify structures, we are using semantic segmentation based on the U-Net architecture to create maps showing the distribution of both chiralities. Additionally, we use a faster R-CNN architecture for object detection to locate and classify each chiral structure individually.
Since the approaches serve different use cases, they also have separate label requirements. We show that models trained using only our synthetic images perform successfully for real STM Images.