6a. Nuclear Forensics Scanning

Nuclear forensics analysis involves the scanning of potentially contaminated material samples with high resolution microscopy to search for evidence of nuclear activity. An electron microscope can resolve details with resolutions on the order of nanometers, but it can take several days to image a one-centimeter squared region of interest at nanometer resolution. A need exists to reduce the required imaging time for effective material sample analysis. What is needed to help nuclear forensics analysis is a software architecture for reducing the imaging time in nuclear forensics by automating material sample scans with electron microscopes. The proposed software would perform a lower-resolution surface scan over the area of interest, identify anomalies and select a distribution of relevant sample subregions for high resolution imaging. The high-resolution images would then be passed to downstream statistical analysis packages to gather statistical data on the various subregion representations found in the material sample. In Phase I, relevant deep learning techniques in anomaly detection, classification, and segmentation will be unified to develop a software architecture for automation of critical processes in material sample scanning. The system’s architecture will be studied and refined in consultation with experts in materials science and electron microscopy. This will involve the acquisition of material samples and development of simulated sample data for the training, validation, and testing of the final deep learning architectures. The proposed software tool will then be tested on a previously analyzed sample to demonstrate its efficacy over human-based detection processes. Nuclear forensics would benefit because the cost of scanning samples would be reduced. Biological and materials sciences can also benefit from automation techniques in microscopy. The ability to efficiently scan material samples in hours rather than days or months, if not years, to acquire important statistics would help revolutionize basic research in systems biology as well as materials science. The ability to efficiently correlate microstructures to a larger sample would aid researchers in targeting their collective efforts to areas of significance and help drive basic research in areas such as pathology and oncology.