Uranium Oxide Synthetic Pathway Discernment through Unsupervised Morphological Analysis

We present a novel unsupervised machine learning method for quantitative representation of scanning electron micrographs and its applications performance nuclear forensic analysis uranium ore concentrates. The uses vector quantizing variational autoencoder followed by histogram operation to encode micrograph into single dimensional representation, called the feature vector. requires no extant labeling data can be applied over large datasets with minimal human interaction. representations generated are broadly descriptive each microstructure material imaged. In case concentrate analysis, were amenable processing reagent species classification accuracy 81.8%, which is competitive state-of-the-art supervised networks [1]. also able classify previously routes not included in training set, imaging parameters such as magnification (to 76.0% accuracy), fine grained process calcining temperature 74.4% their informatic properties indicate that they generally image represented. This across fields perform without need labor intensive possibly biased analysis.