October 9, 2023
Mary Ward House Conference and Exhibition Centre
Europe/London timezone
Workshop Registration will close at the end of Monday 2nd October

Spectroscopic Anomaly Detection and Isotope Identification using Non-negative Matrix Factorization and Application to AWE SIGMA Data

Oct 9, 2023, 3:05 PM
20m
Mary Ward House Conference and Exhibition Centre

Mary Ward House Conference and Exhibition Centre

London, UK

Speaker

Stefan Faaland (Lawrence Berkeley National Laboratory)

Description

In order to reliably detect and identify weak radiological/nuclear sources in real-world environments while maintaining low probabilities of false alarm, it is necessary to employ algorithms that are able to account for temporally and spatially varying backgrounds, exploit the full information content of acquired spectra, and provide interpretable detection metrics.

Over the last several years, Lawrence Berkeley National Laboratory has demonstrated the use of Non-negative Matrix Factorization (NMF) [1] as a framework for the analysis of spectroscopic radiation data [2]. Anomaly detection and isotope identification algorithms [3,4] based on the use of NMF have shown the ability to offer state-of-the-art detection performance. Recent innovations include online learning approaches that allow NMF models of the background to be updated using recently acquired data, and a method to automatically extract common spectral signatures from real-world anomalies.

In this presentation, we will discuss NMF-based algorithms for spectral anomaly detection and isotope identification, and their application to the analysis of data recorded during the AWE SIGMA pilots in the context of the NuSec/AWE SIGMA Data Challenge.

Focussing on data recorded with several statically placed 2x4x16” NaI(Tl) detectors, we will describe the radiological background and associated variability observed for each detector in the context of the NMF framework, and show a range of observed anomalies including medical and industrial sources, and weather-induced anomalies. Finally, we will provide a general perspective on the use of the data for algorithm development and evaluation, potential next steps, and ongoing activities associated with the development of NMF-based radiological detection algorithms.

References
[1] Lee, D., Seung, H. Learning the parts of objects by non-negative matrix factorization. Nature 401, 788–791 (1999). https://doi.org/10.1038/44565
[2] M. S. Bandstra, T. H. Y. Joshi, K. J. Bilton, A. Zoglauer and B. J. Quiter, "Modeling Aerial Gamma-Ray Backgrounds Using Non-negative Matrix Factorization," in IEEE Transactions on Nuclear Science, vol. 67, no. 5, pp. 777-790, May 2020, doi: 10.1109/TNS.2020.2978798.
[3] K. J. Bilton et al., "Non-negative Matrix Factorization of Gamma-Ray Spectra for Background Modeling, Detection, and Source Identification," in IEEE Transactions on Nuclear Science, vol. 66, no. 5, pp. 827-837, May 2019, doi: 10.1109/TNS.2019.2907267.
[4] J. Lee et al., "An Ensemble Approach to Computationally Efficient Radiological Anomaly Detection and Isotope Identification," in IEEE Transactions on Nuclear Science, vol. 69, no. 10, pp. 2168-2178, Oct. 2022, doi: 10.1109/TNS.2022.3198906.

Primary authors

Stefan Faaland (Lawrence Berkeley National Laboratory) Dr Mark Bandstra (Lawrence Berkeley National Laboratory) Dr Reynold Cooper (Lawrence Berkeley National Laboratory) Andrew Jones (Lawrence Berkeley National Laboratory) Dr Marco Salathe (Lawrence Berkeley National Laboratory) Dr Brian Quiter (Lawrence Berkeley National Laboratory)

Presentation materials