Physics-driven deep learning enables temporal compressive coherent diffraction imaging

Abstract

Coherent diffraction imaging (CDI), as a lensless imaging technique, can achieve a high-resolution image with intensity and phase information from a diffraction pattern. To capture high-speed and high-spatial-resolution scenes, we propose a temporal compressive CDI system. A two-step algorithm using physics-driven deep-learning networks is developed for multi-frame spectra reconstruction and phase retrieval. Experimental results demonstrate that our system can reconstruct up to eight frames from a snapshot measurement. Our results offer the potential to visualize the dynamic process of molecules with large fields of view and high spatial and temporal resolutions.