Real-time radionuclides detection using artificial intelligence

Filipe André Carvalho Mendes

Key information

Authors:

Filipe André Carvalho Mendes (Filipe André Carvalho Mendes)

Supervisors:

Alberto Manuel Martinho Vale (Alberto Manuel Martinho Vale); Bruno Miguel Soares Gonçalves (Bruno Miguel Soares Gonçalves)

Published in

01/27/2021

Abstract

Gamma-ray spectroscopy is the usual method to identify detected radioactive hot-spots. Classical Gamma spectroscopy involves many phases, longer analysis and usually an expert to reach an identification result. Thus, developing and upgrading the identification systems available has been a challenge for security and defence organisations such as Departments of Homeland Security, Emergency Response Teams, Customs and Border Control. This work proposes an approach using machine learning techniques that is intended to be implemented as an easy to use identification system, meaning that it can be used by anyone without experience in the field. The proposed solution makes use of artificial neural networks to produce a classification to a given spectrum obtained with a \ac{CZT} sensor. The system is trained using simulated data and is then tested with real acquisition spectra. Single and multiple isotope identification on each sample is explored, highlighting the benefits of an implementation of this kind as well as possible improvements. Additionally, an example of a portable application is suggested using a Raspberry Pi. It is noteworthy that the artificial neural networks developed could be implemented in other devices such as a mobile phone with a connection to a detector. This kind of standalone and portable system could be used on site by humans or even by unmanned vehicles such as drones.