Neural Network Approaches for Mobile Spectroscopic Gamma-Ray Source Detection

Kyle J. Bilton,Joseph C. Curtis,Kai Vetter,Mark S. Bandstra,Daniel Hellfeld,Tenzing H. Y. Joshi

doi:10.3390/jne2020018

Abstract

Artificial neural networks (ANNs) for performing spectroscopic gamma-ray source identification have been previously introduced, primarily for applications in controlled laboratory settings. To understand the utility of these methods in scenarios and environments more relevant to nuclear safety and security, this work examines the use of ANNs for mobile detection, which involves highly variable gamma-ray background, low signal-to-noise ratio measurements, and low false alarm rates. Simulated data from a 2” × 4” × 16” NaI(Tl) detector are used in this work for demonstrating these concepts, and the minimum detectable activity (MDA) is used as a performance metric in assessing model performance.In addition to examining simultaneous detection and identification, binary spectral anomaly detection using autoencoders is introduced in this work, and benchmarked using detection methods based on Non-negative Matrix Factorization (NMF) and Principal Component Analysis (PCA). On average, the autoencoder provides a 12% and 23% improvement over NMF- and PCA-based detection methods, respectively. Additionally, source identification using ANNs is extended to leverage temporal dynamics by means of recurrent neural networks, and these time-dependent models outperform their time-independent counterparts by 17% for the analysis examined here. The paper concludes with a discussion on tradeoffs between the ANN-based approaches and the benchmark methods examined here.

Highlights

Two key elements of nuclear safety and security are the ability to detect the presence of radioactive sources and to correctly identify radionuclides
Artificial neural networks (ANNs) [2,3] are one class of methods previously introduced for gamma-ray source identification
The first applications of neural networks for source identification were examined between the early 1990s and 2000s [2,7,8,9,10], with networks that mapped input spectra to the relative amount of known background and sources contained within the spectrum

Summary

Introduction

Two key elements of nuclear safety and security are the ability to detect the presence of radioactive sources and to correctly identify radionuclides. ANNs, referred to as neural networks, are used to determine a function which maps a given gamma-ray spectrum to the types of radionuclides, or lack thereof, that are observed in the spectrum. 1173 and 1332 kev for 60Co and 662 keV for 137Cs) In this context, optimization refers to the process of using a dataset X , split into training and validation subsets, to update model parameters P , generally by some variation of stochastic gradient descent, such that the loss evaluated on X decreases with number of training iterations, or epochs. Stopping is the method of stopping the training process once the validation loss begins to increase for some number of iterations, referred to as the patience. Early stopping with a patience of 10 iterations is used in training each network

Objectives

Methods

Results

Conclusion