Computer vision-based neural networks for radioisotope identification in urban environments
2026-06-30 • Machine Learning
Machine Learning
AI summaryⓘ
The authors developed a machine learning method to identify radioactive sources in busy urban environments, which are hard because backgrounds vary a lot and threat signals are rare. They turned radiation data into special images called waterfall spectrograms that show changes over time, letting neural networks learn better by looking at both spectral and time information. They tested three types of neural networks and found a convolutional neural network performed better than a previous method when allowing for some false alarms. However, when very low false alarms were required, their method did not always do better, suggesting more work is needed.
Radioisotope identificationMachine learningGamma-ray spectroscopyWaterfall spectrogramConvolutional neural network (CNN)Vision transformer (ViT)Multilayer perceptron (MLP)False positive rateRadiological anomaly detectionNon-negative matrix factorization (NMF)
Authors
Masen Bachleda, Peter Lalor
Abstract
Algorithm development for radioisotope identification in mobile urban search scenarios face significant challenges from non-uniform backgrounds, momentary source encounters, and severe class imbalance between rare threat signatures and background measurements. We present a machine learning-based approach to this problem that converts list-mode gamma-ray data into two-dimensional waterfall spectrograms and applies computer vision architectures to the resulting images. Rather than treating waterfalls as conventional images, we employ a representation where consecutive time spectra can form input channels, similar to RGB channels in color images. This representation encodes both spectral and temporal information, enabling neural networks to more effectively learn patterns that distinguish source signatures from background fluctuations. We evaluate three architectures, a multilayer perceptron (MLP), convolutional neural network (CNN), and vision transformer (ViT), on the Radiological Anomaly Detection and Identification (RADAI) benchmark dataset. At a false positive rate of less than one false alarm per hour, our CNN outperforms the previous-best non-negative matrix factorization (NMF) method across all global metrics, achieving true detection, classification, and identification rates of 0.4334, 0.3965, and 0.2950 respectively, compared to 0.4151, 0.3611, and 0.2625 for NMF. At lower false positive rate constraints, the neural network approaches show comparable but ultimately lower performance than NMF, indicating opportunities for further research.