Calibration-sample free distortion correction of electron diffraction patterns using deep learning

Abstract

The accuracy of the information that can be extracted from electron diffraction patterns is often limited by the presence of optical distortions. Existing distortion characterization techniques typically require knowledge of the reciprocal lattice of either the sample of interest or a separate calibration sample, the latter of which would need to be swapped in, thus adding time and inconvenience to an experiment. To overcome this limitation, we develop a deep learning (DL) framework for measuring and correcting combinations of different types of optical distortion in CBED patterns. Quantitative performance tests of our DL model are conducted using testing datasets of artificial distorted CBED patterns of molybdenum disulfide on amorphous carbon, with varying sizes of CBED disks, that are generated using multislice simulations. The performance test results of our DL approach are benchmarked against those obtained using a conventional distortion estimation technique that uses the radial gradient maximization (RGM) technique and knowledge of the reciprocal lattice system. While the RGM approach outperforms our DL approach for the CBED patterns with very small disks, our DL approach outperforms the RGM approach for the CBED patterns with medium-sized disks, as well as those with large overlapping disks. The benchmarking results suggest that our DL approach, which does not require knowledge of the sample, achieves a good compromise between convenience and accuracy. We also show how our DL framework can be used to improve experimental ptychographic reconstructions, and to correct optical distortion in experimental selected area electron diffraction patterns.