Fig. 4: Deep-learning-based algorithm for atomic site classification.
From: Recent advances and applications of deep learning methods in materials science
a Deep neural networks U-Net model constructed for quantification analysis of annular dark-field in the scanning transmission electron microscope (ADF-STEM) image of V-WSe2. b Examples of training dataset for deep learning of atom segmentation model for five different species. c Pixel-level accuracy of the atom segmentation model as a function of training epoch. d Measurement accuracy of the segmentation model compared with human-based measurements. Scale bars are 1 nm [Reprinted according to the terms of the CC-BY license ref. 228].
