Regensburg 2019 – scientific programme

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

MA: Fachverband Magnetismus

MA 15: Magnetism Poster A

MA 15.31: Poster

Tuesday, April 2, 2019, 10:00–13:00, Poster E

Towards machine-learning-based far-field phase-retrieval for dichroic imaging — •Michael Lohmann, Ofer Kfir, Sergey Zayko, and Claus Ropers — Georg August Universität Göttingen

Machine learning (ML) is a highly powerful tool for data analysis and classification, across many fields and platforms. First demonstrations of ML for lensless imaging by phase-retrieval of diffraction patterns show great potential [1], however, they do not necessarily provide for a consistent solution, and require a vast amount of training data.
Here, we propose the use of a known forward operator, linking the sample and its diffraction, that is, the Fourier transform, to improve the image retrieval. The resulting reduction of complexity could relieve the need for a large training set, and could reduce the algorithm convergence time. In the case of dichroic imaging, such as in magnetic circular dichroism, advanced algorithms can jointly solve the two dichroic diffraction patterns, and directly access the magnetic information separately from the non-magnetic background. Furthermore, combining magnetic imaging with standard ML applications, as de-noising, would enhance the image quality and sensitivity.

[1] Mathew J. Cherukara, Youssef S. G. Nashed & Ross J. Harder; Scientific Reports 8 16520 (2018), Real-time coherent diffraction inversion using deep generative networks