Radionuclide identification for plastic scintillation detectors by backward cumulative channel sum

Abstract Large-area plastic scintillation detectors are widely used in radiation portal monitors, but their Compton-dominated spectra provide limited radionuclide-specific information. We evaluated backward cumulative channel sum (BCCS) combined with non-negative least-squares (NNLS) decomposition and compared it with raw-spectrum NNLS using an explicit background template and detector-specific five-window NNLS. A measured background and fixed reference spectra for Am-241, Cs-137, and Co-60 were used to analyze 18 independent spectra: nine single-radionuclide measurements, eight binary mixtures, and one ternary mixture. The mixed-source compositions were concealed during the primary BCCS analysis. All three methods ranked the actual radionuclide first for every single-radionuclide spectrum and the two constituents first and second for every binary mixture; the ternary mixture yielded positive coefficients for all three radionuclides. For the nine mixed-source spectra, method-specific relative residuals were 0.48–1.31% for BCCS-NNLS, 0.88–1.60% for raw-spectrum NNLS, and 0.11–0.50% for five-window NNLS; these values were calculated in different representation spaces and should not be compared as equivalent goodness-of-fit measures. Parametric Poisson resampling supported stable coefficient rankings under counting-statistical variation. These results support BCCS-NNLS as a simple, linear, and interpretable approach for static, unshielded measurements within a fixed radionuclide library, without hardware modification or data-intensive training. The study does not address open-set identification, activity quantification, or formal detection limits.