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

Area of Science:

• Materials Science
• Computational Science
• Optics

Background:

• Electron diffraction patterns are crucial for materials analysis but are often affected by optical distortions.
• Current methods for correcting these distortions require a calibration sample or knowledge of the sample's reciprocal lattice, adding complexity and time.

Purpose of the Study:

• To develop a novel deep learning framework for measuring and correcting multiple optical distortions in electron diffraction patterns.
• To eliminate the need for a separate calibration sample in distortion correction.

Main Methods:

• Developed the Python library EMicroML featuring a deep learning model.
• Trained and tested the DL model on artificially distorted convergent beam electron diffraction (CBED) patterns of MoS2 on amorphous carbon using multislice simulations.
• Benchmarked the DL approach against the radial gradient maximization (RGM) technique.

Main Results:

• The DL approach effectively corrects pincushion, spiral, elliptical, and parabolic distortions without sample knowledge.
• The DL method shows superior performance over RGM for medium and large overlapping disks in CBED patterns.
• The DL framework also demonstrated utility in improving ptychographic reconstructions and correcting experimental selected area electron diffraction patterns.

Conclusions:

• EMicroML provides a convenient and accurate solution for optical distortion correction in electron diffraction.
• The deep learning approach offers a significant advancement over traditional methods, especially for complex diffraction patterns.
• This tool enhances the reliability and efficiency of electron microscopy data analysis.