Abstract
The inherently stochastic nature of radiation emissions makes modeling background radiation structure a particularly challenging research area. In source identification scenarios, which are critical to nuclear security, the complexity of background radiation modeling is intensified by dynamically changing factors that influence radiation measurements. Consequently, accurately modeling and estimating background radiation can significantly improve our nuclear security capabilities by enhancing the detection of anomalies within radiation data. This study introduces a new data-driven approach to modeling background radiation from spectral measurements. By leveraging the novel data mining technique, Matrix Profile (MP), this approach identifies structural patterns within radiation measurements. The method was tested on real-world background 1-second spectral data collected across various locations, with results demonstrating MP's effectiveness in modeling background structures for measurements taken in the same location. Additionally, MP modeling outperformed the traditional method of using raw measurement averages, particularly in generating distinct models for low-count backgrounds from different locations.