Semantic Segmentation for Materials Classification of Nuclear Fuels

Abstract

Semantic segmentation, the task of classifying objects in an image at a pixel level, has been done since 2012. While our method is not new, our application is. Unlike most tasks which are on clearly-defined objects, the dataset we attempt to label is like Perlin Noise: seemingly random but with clear patterns throughout. Additionally, we had a very small dataset to work with, but preliminary results show that approaches used on more standard applications also work well in this novel application.