Mode-driven explainable artificial intelligence approach for estimating background radiation spectrum in a measurement applicable to nuclear security

This study introduces an explainable artificial intelligence (XAI) approach designed to estimate background spectra in unknown spectral measurements. The approach combines kernel-modeled Gaussian processes (GP) for naturally occurring radioactive material (NORM) estimation with fuzzy logic inference for isotopic photopeak identification. Recognizing the diverse interpretations of background radiation, the paper’s objective is to propose a multi-mode driven approach, with each mode implementing a distinct set of fuzzy rules, thus modeling different backgrounds. Importantly, each mode includes rules associated with nuclides expected to be present in specific locations, such as medical isotopes in a hospital setting. A key innovation of the method is the additional step of providing explanations for the estimated contributions that accompany the estimated background spectrum. Results obtained from a range of gamma-ray spectra representing different locations demonstrate the framework’s potential in estimating background radiation and aiding decisions in the nuclear security domain, particularly for identifying potential nuclear threats in unknown measurements.