Materials Image Informatics Using Deep Learning

Abstract

The growing application of data-driven analytics in materials science has led to the emergence and popularity of the relatively new field of materials informatics. Of the many types of available data in materials science, image data is quite common and heterogeneous in itself, thanks to the advances in various materials imaging techniques. Within the arena of data analytics techniques, deep learning has recently led to groundbreaking advances in numerous fields such as computer vision. In this chapter, we describe the basics of deep learning, its advantages, challenges, and illustrative applications on materials images at different length scales for the purpose of fast and accurate structure characterization. While it is possible to build an accurate deep learning model from scratch when big data is available, transfer learning is used for small datasets. Together, the advances in materials imaging and deep learning provide unprecedented opportunities for such materials image informatics to enable better and faster understanding of the structure-property relationships in materials science and engineering.