Intelligent Analysis of Low-Count Scintillation Spectra Using Support Vector Regression and Fuzzy Logic

Radioisotope identification from low-count-rate spectra or spectra obtained through low-resolution detectors constitutes a challenging environment for accurate spectral analysis. The use of intelligent processing algorithms is a significant step in analyzing spectra, conceivably increasing the accuracy of the nuclide identification in such scenarios. This paper introduces an intelligent methodology for automated processing of low-count gamma-ray spectra acquired with a scintillation detector aimed at identifying radioisotope patterns, and it evaluates the performance of this methodology against a set of experimentally acquired gamma-ray spectra. The novel methodology adopts tools from the “artificial intelligence library” to preprocess the spectrum and subsequently identify radioisotopes. In particular, in the preprocessing step, the measured spectrum is divided into equally long energy intervals, whose values are replaced with those computed by a support vector regressor equipped with a linear kernel function. In the next step, the fuzzy logic–based identifier matches spectral peaks with entries in the spectral library, aiming at identifying isotopic signatures in the spectrum. The proposed intelligent methodology is benchmarked against the multiple-linear-regression (MLR) spectrum-fitting algorithm. Assessment results demonstrate the effectiveness of the proposed methodology in identifying isotopes compared with the MLR algorithm by significantly reducing the number of false detections and improving correct detection performance. Furthermore, the proposed methodology exhibits an overall higher detection sensitivity (by 13.3%) and precision (by 46.8%) than those obtained with MLR.