The Advanced Radiation Detection, Imaging, Data Science, and Applications (ARDIDSA) group has developed the world’s first public benchmark dataset for the testing and evaluation of radiation detection and identification algorithms in an illicit radioactive source search campaign setting. This dataset includes thousands of synthetic list mode data files representing the response of a mobile 2”x4”x16” NaI(Tl) detector driving through a mid-sized urban city. These data were generated using a state-of-the-art Monte Carlo procedure developed over the course of several years at ORNL that has been rigorously tested and validated using real world data. Included in the data are highly dynamic radiation background levels arising from the variations in the absolute and relative concentrations of K-40, U-235, U-238, Th-232, and their progeny in different materials throughout the simulated city. These variations are a ubiquitous challenge for radiation detection and identification algorithms, making this dataset an invaluable resource for both testing/evaluation of existing algorithms and the development of new ones. A total of six different illicit sources are simulated in the dataset, present at 15 different locations throughout the simulated city. This dataset was recently used in an international public data competition, where the winning algorithms resulted in the development of a variety of new methodologies for performing radioactive source search campaigns. A detailed description of this dataset, along with instructions on how to access it can be found in a peer-reviewed journal article recently published in Natural Publishing Group’s Scientific Data journal, titled “Data for training and testing radiation detection algorithms in an urban environment”: https://www.nature.com/articles/s41597-020-00672-2.
