aalto1 untyped-item.component.html
Informed automated structure discovery of atomic force microscopy image
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Chemical Engineering |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
Department
Major/Subject
Mcode
Language
en
Pages
76
Series
Abstract
Atomic Force Microscopy (AFM) enables direct imaging of atomic-level features however, the interpretation of non-planar molecules is challenging due to the fact that only the top layers of these systems interact with the microscope tip. This leads to images deviating from structures familiar to us. Recent Advances in interatomic potentials and machine learning-based image recognition tools have provided a framework suited to tackle this challenge. However, machine learning methods rely heavily on training data and may produce inaccurate results when faced with unfamiliar structures. An alternative approach is to develop an iterative algorithm that generates realistic 3D structures by comparing simulated and experimental AFM images in a fully automated manner. This work examines various image quality metrics on AFM images to quantitatively compare simulated and experimental images, as well as evaluates feature detection and matching methods to determine the step when the simulated structure closely aligns with experimental data. These methods aim to guide the iterative refinement process, mimicking how an expert would progressively enhance structural accuracy removing the necessity for human oversight. Next, image registration is explored to further align simulation images with experimental references, bringing the underlying configurations closer together. In the last section to ensure physically realistic transitions between structures, minima hopping is applied. These approaches are tested on water clusters modeled on gold and copper surfaces using the Neural equivariant interatomic potential (NequIP).
Description
Supervisor
Foster, AdamThesis advisor
Jestilä, JoakimStark, Robert