Application of deep metric learning model to microscope image analysis for the determination of UOC samples in nuclear forensics analysis

This study discusses the application of a deep metric learning model based on a convolutional neural network to scanning electron microscope image analysis to determine UOC samples. One of the unique features of this technique is that it can detect a sample that comes from an unknown material not listed in the reference for comparison, in addition to the classification of a sample based on surface characteristics captured in the microscopic images. It was confirmed that the present technique could detect hypothetical unknown samples with > 0.8 of Area Under the ROC Curve, and it can effectively provide preliminary observations in nuclear forensics analysis.