Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 3: Poster
AKPIK 3.7: Poster
Donnerstag, 21. März 2024, 11:00–14:30, Poster B
Phase retrieval by a conditional Wavelet Flow: applications to near-field X-ray holography — •Ritz Aguilar1, Yunfan Zhang1, Anna Willmann1, Erik Thiessenhusen1, Johannes Dora2, Johannes Hagemann2, Andre Lopes3, Imke Greving3, Berit Zeller-Plumhoff3, Markus Osenberg4, Michael Bussmann1, and Jeffrey Kelling1 — 1Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany — 2Deutsches Elektronen-Synchrotron, Hamburg, Germany — 3Helmholtz-Zentrum Hereon, Geesthacht, Germany — 4Helmholtz-Zentrum Berlin, Berlin, Germany
Phase retrieval (PR) is an ill-posed inverse problem with several applications in medical imaging and materials science. Conventional PR algorithms either simplify the problem by assuming certain object properties and optical propagation regimes or tuning a large number of free parameters which is a time-consuming process. To circumvent this, a machine learning algorithm based on normalising flows (NF) can be used for good inversion, efficient sampling, and fast density estimation of complex-valued distributions. Here, complex wavefield data are trained on a NF-based model called conditional Wavelet Flow (cWF) which adds a conditioning network on top of the Wavelet Flow algorithm. It directly models the conditional data distribution of high resolution images and takes advantage of the parallelized training of different image resolutions, allowing for faster training of large datasets. The trained cWF is then applied to near-field X-ray holography data wherein fast and high-quality image reconstruction is made possible.
Keywords: phase retrieval; machine learning; normalising flows; conditional wavelet flow; near-field holography