Publication date: 15th May 2025
Understanding the morphology of nanoparticles is essential for controlling catalytic activity, stability, and electronic properties. Traditional ensemble measurements often fail to capture fine-scale morphological variations, especially in heterogeneous nanoparticle populations. Capturing these details requires individual particle analysis across large datasets. To address this, we previously developed a high-throughput pipeline leveraging deep learning for image segmentation and shape quantification of individual nanoparticles from high-resolution transmission electron microscopy (HRTEM) images. This approach enables subnanometer (~0.2 nm) resolution, allowing us to correlate size and shape features in statistically meaningful ways across large particle populations.
In this presentation, we focus on the synthesis of cubic-shaped cobalt oxide (Co3O4) nanoparticles, systematically varying synthetic parameters such as cobalt precursor concentration and water content to control size and shape descriptors such as circularity and face convexity. We acquire HRTEM images of hundreds of thousands of these nanoparticles under various synthetic conditions, encompassing more than 441,000 particles in total. Conventional computer vision methods often struggle with such high-resolution images. Therefore, we employ convolutional neural networks (CNNs) for pixel-level segmentation, enabling precise characterization of individual particles. This machine learning-assisted, population-wide statistical analysis reveals size-dependent morphological transitions and provides critical insights into the growth mechanisms of nanoparticles. Specifically, we uncover growth regime transitions, including the evolution from convex to concave polyhedral shapes. We also introduce the concept of an “onset radius,” describing critical thresholds for these transitions.