Data Science in Support of Radiation Detection for Border Monitoring: An Exploratory Study

Abstract

Abstract Radiation detection technology is widely deployed to identify undeclared nuclear or radiological materials in transit. However, in certain environments the effective use of radiation detection systems is complicated by the presence of significant quantities of naturally occurring radioactive materials that trigger nuisance alarms which divert attention from valid investigations. The frequency of nuisance alarms sometimes results in the raising of alarming thresholds, reducing the likelihood that systems will detect the low levels of radioactivity produced by key threat materials such as shielded highly enriched uranium. This paper explores the potential of using data science techniques, such as dynamic time warping and agglomerative hierarchical clustering, to provide new insights into the cause of alarms within the maritime shipping environment. These methods are used to analyze the spatial radiation profiles generated by shipments of naturally occurring radioactive materials as they are passed through radiation portal monitors. Applied to a real-life dataset of alarming occupancies, the application of these techniques is shown to preferentially group and identify similar commodities. With further testing and development, the data-driven approach to alarm assessment presented in this paper could be used to characterize shipments of naturally occurring radioactive materials at the primary scanning stage, significantly reducing time spent resolving nuisance alarms.