Unsupervised domain adaptation for radioisotope identification in gamma spectroscopy

TL;DR

This paper demonstrates that unsupervised domain adaptation techniques can significantly improve the generalization of radioisotope identification models trained on synthetic gamma spectroscopy data when applied to real-world data.

Contribution

It compares various UDA methods and shows that feature alignment strategies like MMD minimization enhance model performance in practical scenarios.

Findings

Unsupervised feature alignment improves testing accuracy from 0.754 to 0.904.

MMD minimization and domain-adversarial training are effective UDA strategies.

Models trained on synthetic data can be adapted for real-world use with UDA.

Abstract

Training machine learning models for radioisotope identification using gamma spectroscopy remains an elusive challenge for many practical applications, largely stemming from the difficulty of acquiring and labeling large, diverse experimental datasets. Simulations can mitigate this challenge, but the accuracy of models trained on simulated data can deteriorate substantially when deployed to an out-of-distribution operational environment. In this study, we demonstrate that unsupervised domain adaptation (UDA) can improve the ability of a model trained on synthetic data to generalize to a new testing domain, provided unlabeled data from the target domain is available. Conventional supervised techniques are unable to utilize this data because the absence of isotope labels precludes defining a supervised classification loss. We compare a range of different UDA techniques, finding that feature alignment strategies, particularly via maximum mean discrepancy (MMD) minimization or domain-adversarial training, yield the most consistent improvement to testing scores. For instance, using a custom transformer-based neural network, we achieve a testing accuracy of $0.904 \pm 0.022$ on an experimental LaBr$_3$ test set after performing unsupervised feature alignment via MMD minimization, compared to $0.754 \pm 0.014$ before alignment. Overall, our results highlight the potential of using UDA to adapt a radioisotope classifier trained on synthetic data for real-world deployment.