Mainz 2026 – scientific programme

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

MO: Fachverband Molekülphysik

MO 28: Interaction with VUV and X-ray light I (joint session A/MO)

MO 28.4: Talk

Thursday, March 5, 2026, 15:30–15:45, N 2

Neural networks for diffraction image separation — •Nikita Morozov — European XFEL, Schenefeld, Germany

Newly available soft X-ray two-color FEL pulse mode at European XFEL opens a new path to the structural and plasma studies at the nanometer scale. The first pulse, designated as the pump, characterizes the initial state of the object, whereas the second probe pulse captures the system's evolution following its interaction with the pump. The pulses are separated by a short time interval of less than 1 ps. Because this delay is shorter than the detector acquisition time, the state-of-the-art detectors are unable to capture two independent Coherent Diffractive Imaging (CDI) images corresponding to the pump and probe pulses. Instead, it records a single data frame that contains a superposition of CDIs from both FEL pulses, making it almost impossible to follow the evolution process.

This work presents machine learning-based solutions for separating the overlapping components in such data sets. We propose two methods: the first one is based on diffusion probabilistic models, a recent and powerful approach for image generation, and the second using feed-forward convolutional neural networks to solve the same task.

Ideally, after the image separation task is done, the pump and probe densities could be recovered using standard phase retrieval techniques, allowing the excitation-induced changes to be examined in a time-resolved manner.

Keywords: Coherent Diffractive imaging; machine learning; X-Ray pump-probe; plasma studies; structural changes