Modified UNet++ with attention gate for graphene identification by optical microscopy

Abstract

Graphene has attracted a lot of interest since its discovery. However, graphene layers made by mechanical exfoliation need to be carefully distinguished from multi-layer graphite and residues by experienced experts, which is time consuming and requires significant experience. In this paper, an image segmentation method based on deep learning is developed to identify single-layer graphene (SLG) under an optical microscope. By introducing a modified UNet++ with an attention gate and a residue network (ResNet) for further classification as a two-level structure, we can distinguish SLG from graphite with high accuracy by using only a small amount of training images. The high accuracy of SLG identification and the short inference time make it a promising real-time detection tool besides traditional and technically more involved identification methods such as Raman spectroscopy and atomic force microscopy.