Classifying optical microscope images of exfoliated graphene flakes by data-driven machine learning

Abstract

Machine-learning techniques enable recognition of a wide range of images, complementing human intelligence. Since the advent of exfoliated graphene on SiO₂/Si substrates, identification of graphene has relied on imaging by optical microscopy. Here, we develop a data-driven clustering analysis method to automatically identify the position, shape, and thickness of graphene flakes from optical microscope images of exfoliated graphene on an SiO₂/Si substrate. Application of the extraction algorithm to optical images yielded optical and morphology feature values for the regions surrounded by the flake edges. The feature values formed discrete clusters in the optical feature space, which were derived from 1-, 2-, 3-, and 4-layer graphene. The cluster centers are detected by the unsupervised machine-learning algorithm, enabling highly accurate classification of monolayer, bilayer, and trilayer graphene. The analysis can be applied to a range of substrates with differing SiO₂ thicknesses.

Machine learning: microscope classification of graphene

Graphene layers can be detected on a substrate with high accuracy using a machine-learning algorithm. A team led by Tomoki Machida at the University of Tokyo developed a method based on unsupervised data-driven clustering analysis to identify and classify exfoliated graphene flakes on a SiO₂/Si substrate. The algorithm could automatically identify the positions, shapes, and thickness of graphene flakes from a large amount of optical microscope images. Processing of 7 × 10⁴ images of SiO₂/Si substrates yielded 4 × 10⁵ regions enclosed by edges, and further analysis of these segmented areas indicated the presence of N-layered graphene, where N is the number of layers, with >95% accuracy. Integration of the current algorithm with optical microscopes would allow the development of a fully automated machine for identification of graphene flakes.