Convolutional Neural Networks for Challenges in Automated Nuclide Identification.

Anthony Turner,Carl Wheldon,Tzany Wheldon,Lee Packer,Jonathan Burns,Martin Freer,Mark Gilbert

doi:10.3390/s21155238

Abstract

Improvements in Radio-Isotope IDentification (RIID) algorithms have seen a resurgence in interest with the increased accessibility of machine learning models. Convolutional Neural Network (CNN)-based models have been developed to identify arbitrary mixtures of unstable nuclides from gamma spectra. In service of this, methods for the simulation and pre-processing of training data were also developed. The implementation of 1D multi-class, multi-label CNNs demonstrated good generalisation to real spectra with poor statistics and significant gain shifts. It is also shown that even basic CNN architectures prove reliable for RIID under the challenging conditions of heavy shielding and close source geometries, and may be extended to generalised solutions for pragmatic RIID.

Highlights

The performance of any Radio-Isotope IDentification (RIID) model based on machine learning is dependent on two primary aspects
Including as much real data as possible is, highly recommended, whatever the application, and these results clearly demonstrate the advantages of doing so for the purposes of RIID
This work demonstrates that even basic Convolutional Neural Network (CNN) may be used for the rapid, accurate identification of multi-isotope sources

Summary

Introduction

Radio-Isotope IDentification (RIID) finds a broad scope of applications in security, decommissioning, and public health and safety. Each come with their own challenges, but the effectiveness of any detection system stems from its physical limitations and algorithm performance. Persistent challenges in the explicit handling of transient effects (such as gain drift) have led to new approaches with machine learning [1,2,3,4,5,6] By definition, these models operate without being explicitly programmed, instead making decisions based on generalised rules born of experience

Methods

Results

Conclusion