Mask R-CNN to Classify Chemical Compounds in Nanostructured Materials

Abstract

Nowadays artificial intelligence has become the iron horse in high-performance computing, solving problems that were impossible 10 years ago. This work uses a deep learning technique named Mask Region-Convolutional Neural Network (Mask R-CNN) using images of nanostructured materials obtained from a transmission electron microscope (TEM) at a nanoscale spatial resolution. In those images, we observed different regions with specific structure correspond to yttrium silicate and silicon oxide materials system. The architecture Mask R-CNN was trained with TEM images, and performs the classification, location, and segmentation of chemical compounds with a data set of 26 images, reaching scores above 90% of accuracy.