• machine learning
• spectroscopy
• gamma-ray
• pattern recognition
• classification

The identification of radioisotopes remains a challenging task due to several aspects such as low concentration, environmental factors, source-detector distance variations. The aim of this research is the application of three machine learning algorithms (fully connected neural network, convolutional neural network and random forest) to investigate the above-mentioned problems. The machine learning algorithms are trained on a gamma-spectrum template provided by GADRAS-DRF software. Other features to be discussed include the recognition and the identification of correct region-of-interest for the analysis, by exploiting the pattern recognition ability of neural networks. The results presented in this thesis show that Machine Learning is able to identify radionuclides inside single and composed spectra.