End-to-End Automated Segmentation Framework for Four-Dimensional Scanning Transmission Electron Microscopy Data.

Area of Science:

• Materials Science
• Nanotechnology
• Data Analysis

Background:

• High-throughput synthesis produces large nanoparticle arrays.
• Four-dimensional scanning transmission electron microscopy (4D-STEM) generates massive datasets.
• Efficient analysis of 4D-STEM data is crucial for characterizing nanoparticles.

Purpose of the Study:

• To develop an automated framework for segmenting nanoparticles in 4D-STEM data.
• To identify regions with distinct crystal grain composition and orientation.
• To enable efficient compression of large 4D-STEM datasets.

Main Methods:

• An end-to-end segmentation framework using 4D-STEM data.
• Physics-informed extraction of Bragg disk information.
• Feature vector creation combining diffraction and real-space data.
• Gaussian Mixture Model (GMM) for segmentation.
• Development of visualization tools for interface transition and superposition.

Main Results:

• Successful automatic segmentation of nanoparticles into distinct crystal grain regions.
• Identification of interface transitions and degree of superposition.
• Demonstrated capability to compress large 4D-STEM datasets by replacing full data with grain-specific features.
• Validation on real, complex nanoparticle datasets.

Conclusions:

• The proposed framework effectively segments nanoparticles in 4D-STEM data.
• It integrates machine learning with physics knowledge for comprehensive analysis.
• This approach offers significant data compression for large-scale nanoparticle characterization.