Topic:
An unsupervised machine learning algorithm is developed for detecting crystallographic defects in atomic-resolution images. The method enables fast screening of large volumes of atomic-resolution data of a variety of different crystal structures without needing manually labeled training images.
A scanning transmission electron microscope (STEM) is used to obtain atomic-resolution images of various crystalline materials. A Patterson function is used to transform segments of each image to make them invariant to translation. A one-class support vector machine is used to detect defects which are treated as outliers. DOI: 10.1038/s41524-021-00642-1
