Understanding the geometry of nanomaterials at the atomic scale provides critical insights into local structural heterogeneities and their impact on functional properties. Since shapes vary from particle to particle, detailed analysis at the single-particle level is essential. In this talk, I will present a high-throughput pipeline that integrates deep learning-based segmentation with quantitative shape analysis of individual nanoparticles from high-resolution transmission electron microscopy (HRTEM) images. First, I will describe the application of convolutional neural networks (CNNs) to segment 727 HRTEM micrographs of cubic Co
3O
4 nanoparticles, enabling the extraction of shape descriptors from 441,067 particles. This automated workflow allows for population-wide statistical characterization, bridging local structural detail with large-scale analysis. Second, I will present a size-resolved shape analysis at subnanometer precision, highlighting a critical threshold, “onset radius”, that marks transitions in particle shape, such as surface faceting and a shift from thermodynamic to kinetic growth regimes. This bottom-up approach illustrates how machine learning and data-driven analysis can reveal previously unquantified trends, offering a generalizable framework for high-throughput materials characterization.
[more]
