Abstract Scope |
Transmission electron microscopy (TEM) is an indispensable methodology to characterize materials structures and compositions at the atomic scale. Recent advancement of in-situ TEM techniques enables unprecedented capabilities for spatiotemporal and multi-modal characterizations of dynamic material transformations during physical and chemical processes, thereby generating massive high-dimensional datasets. It is highly desired that such large volumes of data can be efficiently and effectively analyzed to accelerate scientific discovery. With rapid growth of artificial intelligence (AI), machine learning (ML) approaches have demonstrated the unique power in analyzing and processing microscopy-related datasets, such as time-sequential and hyperspectral images. Herein, we present our recent efforts in ML-enhanced data analytics to enable high-throughput image processing with less human perception. These methods, demonstrated in atomic-resolution crystal lattices and nanoscale particle systems, allow us to perform reliable and reproducible analysis of TEM-based datasets in multi-modalities and multi-dimensionalities. |