Group 6: Challenge: Task 3 - detect defects in electron microscopy images

Graphene is an exotic nanomaterial consisting of a single layer of carbon atoms arranged in a two-dimensional honeycomb lattice (Figure 1). Since its rediscovery and isolation from graphite by tape in 2009, [1] graphene has aroused intensive interest in the worldwide and in numerous research areas, due to its extraordinary mechanical, thermal, optical, electronic, and other properties. The discoverer Geim and Novoselov were awarded the Nobel Prize in Physics in 2010. Although graphene samples with high perfection have such outstanding performance, the defects in graphene are inevitable during the growth and processing, and can significantly affect the properties. Thus, it is important to figure out the amount, position, size and type of defects in a graphene sample. Machine learning, an emerging data-driven approach, offers a highly efficient solution to learning hidden patterns or classifying anomalies from complex data. In this work, we applied machine learning to detect the defects in the electronic microscopic images of graphene samples and compared the performance of several different machine learning algorithms. As the data set contained significantly fewer defect examples than perfect ones, only methods suitable for use with imbalanced data were considered