Encoding PGAA spectra as images for material classification with convolutional neural networks

We trained deep convolutional neural networks (CNN) to classify a material based on its prompt gamma ray activation analysis (PGAA) spectrum. We focused on two dimensional (2D) models to leverage abundant open-source models pre-trained on other computer vision tasks for transfer learning. This allows models to be built with a relatively small number of trainable parameters. Moreover, CNNs can be explained naturally using class activation maps and can be equipped with out-of-distribution tests to identify materials which were not present in its training set. Together, these features suggest such models may be excellent candidates for automated material identification in real-world scenarios.