Radioactive hot-spot detection using machine learning algorithms

Miguel Tomé Ramos e Barros

Key information

Authors:

Miguel Tomé Ramos e Barros (Miguel Tomé Ramos e Barros)

Supervisors:

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

Published in

01/27/2021

Abstract

The detection of radioactive hot-spots has been a challenge for the security sector, especially in situations involving chemical, biological, radiological and nuclear (CBRN) threats. This work proposes a solution based on Machine Learning techniques, with a focus on Artificial Neural Networks (ANNs), so observations of radiological intensity counts and corresponding localizations can be used to estimate not only the number of unknown radioactive sources present in a given scenario, but also their location and activity at the same time. For this, a simulator is used to generate a training data set for the training process, and so, using the model already trained through a Divide and Conquer algorithm, fast and accurate predictions are achieved, ensuring the reliability of such an ANN-based approach. The proposed solution is then tested in scenarios with multiple sources, with obstacles included, and with non-point-like sources. Unlike most existing algorithms, which begin failing in scenarios with those conditions, ANNs have shown that are capable of performing an accurate hot-spots detection, with a low number of limitations. Additionally, experimental results, done in lab environments and real scenarios located at old deposits of radioactive ore, have shown that the algorithm is scalable for very large regions, as well as for very short scenarios. Thus, ANNs have demonstrated the capability of being an emerging tool with the potential to make a difference in the nuclear field, by helping in the development of novel techniques and new solutions in order to safeguard human lives.