Radiological source search is a challenge involving the detection, identification, and localization of weak sources within background environments that vary both spatially and temporally. In this work, a method for simultaneously detecting and identifying gamma-ray sources using background models formed from spectral data is described. Non-negative matrix factorization (NMF) is used to generate low-dimensional representations of gamma-ray spectra, allowing for a compact means of capturing variation in gamma-ray backgrounds. Background models formed using NMF are used to perform anomaly detection, and additionally, models are augmented with spectral templates of gamma-ray sources to perform simultaneous detection and identification using a likelihood ratio test. The NMF-based detection and identification algorithm is benchmarked against a standard Region of Interest algorithm and shows significant performance gains. Additionally, NMF-based anomaly detection shows improvements over methods based on gross counts or Principal Component Analysis. Algorithm performance is evaluated using unshielded sources with activities between 5 μCi and 400 μCi at a standoff distance of 20 m using source injection on background data collected using a 1 m2 NaI array on the Radiological Multi-sensor Analysis Platform.

As more complex predictive models are used for gamma-ray spectral analysis, methods are needed to probe and understand their predictions and behavior. Recent work has begun to bring the latest techniques from the field of Explainable Artificial Intelligence (XAI) into the applications of gamma-ray spectroscopy, including the introduction of gradient-based methods like saliency mapping and Gradient-weighted Class Activation Mapping (Grad-CAM), and black box methods like Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). In addition, new sources of synthetic radiological data are becoming available, and these new data sets present opportunities to train models using more data than ever before. In this work, we use a neural network model trained on synthetic NaI(Tl) urban search data to compare some of these explanation methods and identify modifications that need to be applied to adapt the methods to gamma-ray spectral data. We find that the black box methods LIME and SHAP are especially accurate in their results, and recommend SHAP since it requires little hyperparameter tuning. We also propose and demonstrate a technique for generating counterfactual explanations using orthogonal projections of LIME and SHAP explanations.

Artificial neural networks (ANNs) for performing spectroscopic gamma-ray source identification have been previously introduced, primarily for applications in controlled laboratory settings. To understand the utility of these methods in scenarios and environments more relevant to nuclear safety and security, this work examines the use of ANNs for mobile detection, which involves highly variable gamma-ray background, low signal-to-noise ratio measurements, and low false alarm rates. Simulated data from a 2” × 4” × 16” NaI(Tl) detector are used in this work for demonstrating these concepts, and the minimum detectable activity (MDA) is used as a performance metric in assessing model performance.In addition to examining simultaneous detection and identification, binary spectral anomaly detection using autoencoders is introduced in this work, and benchmarked using detection methods based on Non-negative Matrix Factorization (NMF) and Principal Component Analysis (PCA). On average, the autoencoder provides a 12% and 23% improvement over NMF- and PCA-based detection methods, respectively. Additionally, source identification using ANNs is extended to leverage temporal dynamics by means of recurrent neural networks, and these time-dependent models outperform their time-independent counterparts by 17% for the analysis examined here. The paper concludes with a discussion on tradeoffs between the ANN-based approaches and the benchmark methods examined here.

Networked detector systems can be deployed in urban environments to aid in the detection and localization of radiological and/or nuclear material. However, effectively responding to and interpreting a radiological alarm using spectroscopic data alone may be hampered by a lack of situational awareness, particularly in complex environments. This study investigates the use of LiDAR and streaming video to enable real-time object detection and tracking, and the fusion of this tracking information with radiological data for the purposes of enhanced situational awareness and increased detection sensitivity. This work presents an object detection, tracking, and novel source-object attribution analysis that is capable of operating in real-time. By implementing this analysis pipeline on a custom developed system that comprises a static 2 × 4 × 16 inch NaI(Tl) detector co-located with a 64-beam LiDAR and 4 monocular cameras, we demonstrate the ability to accurately correlate trajectories from tracked objects to spectroscopic gamma-ray data in real time, and use physics-based models to reliably discriminate between source-carrying and non-source-carrying objects. In this work, we describe our approach in detail and present a quantitative performance assessment that characterizes the source-object attribution capabilities of both video and Li-DAR. Additionally, we demonstrate the ability to simultaneously track pedestrians and vehicles in a mock urban environment, and use this tracking information to improve both detection sensitivity and situational awareness using our contextual-radiological data fusion methodology.