Deep Learning Coherent Diffractive Imaging

TL;DR

This paper introduces a deep learning approach using CNNs trained on simulated data to achieve sub-angstrom resolution in coherent electron diffractive imaging, enabling real-time phase retrieval and broad scientific applications.

Contribution

The study demonstrates that CNNs can replace traditional iterative algorithms for phase retrieval in electron diffraction, achieving comparable resolution with faster, real-time processing.

Findings

Achieved 0.55-0.70 Å spatial resolution.

Successfully applied CNNs to real diffraction data.

Matched quality of conventional ptychographic reconstructions.

Abstract

We report the development of deep learning coherent electron diffractive imaging at sub-angstrom resolution using convolutional neural networks (CNNs) trained with only simulated data. We experimentally demonstrate this method by applying the trained CNNs to directly recover the phase images from electron diffraction patterns of twisted hexagonal boron nitride, monolayer graphene and a Au nanoparticle with comparable quality to those reconstructed by a conventional ptychographic method. Fourier ring correlation between the CNN and ptychographic images indicates the achievement of a spatial resolution in the range of 0.70 and 0.55 angstrom (depending on different resolution criteria). The ability to replace iterative algorithms with CNNs and perform real-time imaging from coherent diffraction patterns is expected to find broad applications in the physical and biological sciences.