Spectral unmixing for activity estimation in Gamma-Ray Spectrometry

Jiaxin Xu (1) , Jerome Bobin (2) , Anne de de Vismes Ott (1) , Christophe Bobin (2)

1 IRSN - Institut de Radioprotection et de Sûreté Nucléaire
2 CEA - Commissariat à l'énergie atomique et aux énergies alternatives

A challenging problem in the domain of gamma-ray spectrum analysis is the rapid detection of artificial radionuclides, which are present at low activity levels. We introduce in this paper new algorithms for activity estimation based on spectral unmixing techniques, which aim to decompose a measured spectrum into individual spectra of radionuclides. We propose to tackle the activity estimation problem as an inverse problem, where individual activities appear as mixing weights related to individual spectra. In contrast to standard approaches, this allows us to account for the full spectrum of each radionuclide (i.e. peaks and Compton continuum). In this article, we investigate different approaches to solve the underlying spectral unmixing problem: standard regularized least squares regression (LS) and a novel regularized maximum likelihood estimation that allows to precisely account for the true Poisson statistics of the physical process underlying the detection. Both methods implement the non-negativity constraint to enforce the fact that the activity of radionuclides cannot be negative. Experimental results on simulated and measured spectra are presented and compared to standard methods, it is shown that the proposed approach leads to more accurate estimations, especially when the counting rate is low, which gives a significant advantage for the rapid detection.

Mots clés: gamma-ray spectroscopy; spectral unmixing; inverse problem; proximal algorithms

Domaines: Sciences de l'ingénieur [physics]; Traitement du signal et de l'image [eess.SP]

Jiaxin Xu, Jerome Bobin, Anne de de Vismes Ott, Christophe Bobin. Spectral unmixing for activity estimation in Gamma-Ray Spectrometry. 2019. ⟨hal-02060476⟩