Radionuclide identification system using convolution neural network for environmental radiation monitoring

Abstract

Radionuclide identification is an important task for nuclear safety and security aspects, especially to environmental radiation monitoring systems. This study aims to build an automatic radionuclide identification system that can be applied in environmental radiation monitoring stations. The gamma energy spectrum was obtained by varying radionuclide types, measurement time and source distance using a scintillation detector. The dataset was collected by converting gamma energy spectrum into images, data pre-processing by removing background noise and normalizing the gamma spectrum. Automatic identification is demonstrated as a development method based on convolutional neural network (CNN) algorithm, where the images come from gamma-ray spectrum in the form of photoelectric peak characteristic. Three CNN architectures are used to train the model, which are VGG-16, AlexNet and Xception. The performance of each model is evaluated using accuracy, precision and recall to find the appropriate architecture. The most optimum results are shown by VGG-16 with an accuracy of 97.72%, a precision of 97.75% and a recall of 97.71%. The models are critically reviewed and it is concluded that the developed models can be further implemented on embedded devices utilizing the tiny machine learning (TinyML) platform in environmental radiation monitoring systems.