Effect of natural gamma background radiation on portal monitor radioisotope unmixing

TL;DR

This study evaluates how natural gamma background radiation affects the accuracy of a radionuclide unmixing algorithm used in radiation portal monitors, demonstrating robustness at higher counts and identifying challenges at lower counts.

Contribution

It provides experimental evidence on the robustness of a state-of-the-art unmixing algorithm against varying background spectra in real-world conditions.

Findings

Algorithm reliably identifies radionuclides with at least 1,000 counts.

Lower counts lead to approximately 36% larger errors in fraction estimation.

Performance remains consistent across different background intensities and spectral shapes.

Abstract

National security relies on several layers of protection. One of the most important is the traffic control at borders and ports that exploits Radiation Portal Monitors (RPMs) to detect and deter potential smuggling attempts. Most portal monitors rely on plastic scintillators to detect gamma rays. Despite their poor energy resolution, their cost effectiveness and the possibility of growing them in large sizes makes them the gamma-ray detector of choice in RPMs. Unmixing algorithms applied to organic scintillator spectra can be used to reliably identify the bare and unshielded radionuclides that triggered an alarm, even with fewer than 1,000 detected counts and in the presence of two or three nuclides at the same time. In this work, we experimentally studied the robustness of a state-of-the-art unmixing algorithm to different radiation background spectra, due to varying atmospheric conditions, in the 16 $^\circ$C to 28 $^\circ$C temperature range. In the presence of background, the algorithm is able to identify the nuclides present in unknown radionuclide mixtures of three nuclides, when at least 1,000 counts from the sources are detected. With fewer counts available, we found larger differences of approximately 35.9$\%$ between estimated nuclide fractions and actual ones. In these low count rate regimes, the uncertainty associated by our algorithm with the identified fractions could be an additional valuable tool to determine whether the identification is reliable or a longer measurement to increase the signal-to-noise ratio is needed. Moreover, the algorithm identification performances are consistent throughout different data sets, with negligible differences in the presence of background types of different intensity and spectral shape.