Next generation particle identification techniques for large geometry secondary ion mass spectrometry (LG-SIMS) analysis of environmental swipe samples for nuclear safeguards

Secondary ion mass spectrometry (SIMS) is a useful microanalytical method to exploit isotopic signatures present in enrichment facilities and assists in determining an estimate of the enrichment levels and operational history via particle analysis. Large geometry (LG) SIMS is particularly well equipped to detect and analyze actinide elements in swipe samples from declared facilities to verify if operations are consistent with safeguards declarations or expectations. For this reason, LG-SIMS has become the primary technique of the International Atomic Energy Agency (IAEA) for fast, accurate and automated particle screening. The efficacy of particle analysis by LG-SIMS is primarily limited by the process of accurately identifying isotopically unique particles of interest (POIs) in the presence of background signatures. In the interest of improving LG-SIMS particle analysis, several novel techniques are being developed in order to better identify regions of interest (ROIs) within LG-SIMS images during particle screening. One of these techniques first developed for small geometry (SG) SIMS involves a marker-controlled watershed approach (a.k.a. Seeker) used to identify ROIs in SIMS image data where there is high variability in image intensity, particles are touching or are in close proximity to one another and/or the total amount of ion signal for a given region is count limited. In this work, an image segmentation method based on the Seeker algorithm has been applied to LG-SIMS images of U-bearing particles. Another method applied in this work builds on the static algorithms inherent in the Seeker approach and is based on machine learning. This method uses an artificial neural network (ANN) that is trained by a skilled SIMS operator to identity ROIs within the SIMS images. Once trained, the ANN can identify isotopically unique POIs without input from the SIMS operator. Over time, more input can be added to the ANN to improve its training and consequently improve the fidelity of the particle searching analysis. The particle identification techniques presented in this work represent the next generation of LG-SIMS particle analysis for nuclear safeguards applications and have the potential to greatly enhance future automated particle searching methodologies.