Rapid and flexible segmentation of electron microscopy data using few-shot machine learning

Automatic segmentation of microstructural features in atomic-scale electron microscope images is crucial for understanding structure-property relationships. Manual analysis is time-consuming, biased, and error-prone. While automated approaches exist, they often lack robustness and generalizability. This paper presents a flexible, semi-supervised few-shot machine learning approach for segmenting scanning transmission electron microscopy (STEM) images of three oxide material systems: SrTiO3/Ge heterostructures, La0.8Sr0.2FeO3 thin films, and MoO3 nanoparticles. The few-shot learning method proves more robust to noise, reconfigurable, and requires less data than conventional methods, enabling rapid image classification and microstructural feature mapping for high-throughput characterization and autonomous microscope platforms.

Sarah Akers, Elizabeth Kautz, Andrea Trevino-Gavito, Matthew Olszta, Bethany E. Matthews, Le Wang, Yingge Du, Steven R. Spurgeon