SAMOP 2023 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

A: Fachverband Atomphysik

A 12: Poster I

A 12.13: Poster

Tuesday, March 7, 2023, 16:30–19:00, Empore Lichthof

Orientation Recovery of Single-shot Scattering Images of Molecules and 3-Dimensional Density Reconstruction using Machine Learning — •Siddhartha Poddar, Ulf Saalmann, and Jan Michael Rost — Nöthnitzer Str. 38, 01187, Dresden, Germany

As an orientation recovery technique of coherent diffractive images at X-ray free-electron lasers for a single molecule, we have applied a pairwise distance learning algorithm to pairs of scattering images with the help of a deep learning network called Siamese Neural Network (SNN) or popularly known as twin network. With this a priori orientation information in the knowledge, now it is possible to successfully reconstruct the 3D electronic density of the corresponding molecule using many tomographic techniques available. So, the implementation of our reconstruction procedure, from 2D images to 3D molecular structure, has two successive steps. First, we train the twin network with a dataset consisting of approximately 100000 scattering images of an asymmetric 4-atomic test molecule and predict the orientations of a new unseen set of images for the same molecule. Second, we align and arrange this small subset of oriented images to compute the overall structure of the molecule using tomography.