Automated Image Analysis for Single-Atom Detection in Catalytic Materials by Transmission Electron Microscopy

Area of Science:

• Materials Science
• Catalysis Science
• Artificial Intelligence in Microscopy

Background:

• Single-atom catalysts (SACs) are crucial for advanced catalysis, but their characterization using transmission electron microscopy (TEM) faces limitations.
• Challenges include poor statistical significance, reproducibility, and interoperability in manual atom identification from TEM images.

Purpose of the Study:

• To develop a deep-learning method for automated atom detection in TEM images.
• To enable high-throughput characterization of single-atom heterogeneous catalysts (SACs).
• To improve the statistical significance and scalability of structure-performance relationship studies in catalysis.

Main Methods:

• A customized deep-learning model was developed for automated atom detection in TEM image analysis.
• The model was tested on platinum single atoms supported on a functionalized carbon material with complex morphology.
• The method was also validated on an iron SAC supported on carbon nitride.

Main Results:

• The deep-learning model successfully detected over 20,000 atomic positions, enabling statistical analysis of catalyst properties.
• Key properties analyzed include surface density, proximity, clustering extent, and dispersion uniformity of metal species.
• The model demonstrated generalizability for single-atom detection on various carbon-related materials.

Conclusions:

• The developed AI approach significantly accelerates TEM image processing for SACs, reducing human bias and improving standardization.
• This method integrates artificial intelligence into routine TEM workflows, paving the way for more robust characterization of frontier catalytic materials.
• The automated analysis facilitates the establishment of reliable structure-performance relationships for nanostructured catalysts.