The use of artificial neural networks in PVT-based radiation portal monitors

Abstract

Polyvinyl toluene (PVT)-based gamma-ray scintillation detectors are cost effective for use in radiation portal monitors (RPMs) applied to screening for illicit radioactive materials at international border crossings. While such systems can provide good sensitivity for detecting the presence of radioactive materials, they have poor spectral resolution that limits their ability to identify the isotopic content of the source of radiation. Without use of spectral information, RPMs cannot distinguish innocent materials that contain low levels of normally occurring radioactive materials (NORM) from special nuclear materials of concern. Thus, to reduce the number of “nuisance” alarms produced in PVT-based RPMs by innocent materials, algorithms that analyze spectra from PVT detectors must be optimized to make use of the limited information contained in their energy spectra. This paper reports the first application of artificial neural networks (ANNs) in such an analysis. This work was performed as a feasibility study whose primary objective was to describe how an ANN-based alarm algorithm can be used to reduce the nuisance/false alarm probability while maintaining high-detection probabilities for radioactive sources of interest. The spectra used in this study were obtained from a limited set of actual PVT-based RPM data, and included cases where simulated spectra were inserted into the measured spectra. This paper also includes an analysis of spectral channel importance and shows evaluations of two methods used to reduce the initial set of energy spectra channels into smaller sets. Although not a comprehensive study, the results of this work show that it is possible to use ANNs successfully to discriminate NORM from other materials for realistic PVT-based RPM spectra. The algorithms described may also have potential application in the analysis of sodium iodide based RPM spectra.