Source Identification using Non-negative Matrix Factorisation on Gamma-ray Spectra

Poisson non-negative matrix factorisation (PNMF) has been identified as a promising method to quantify and reconstruct gamma-ray spectra for detecting radioactive material using deployed scintillator detectors. Identifying threats represents an important goal to protect countries from illicit radioactive material being used for acts of nuclear terrorism. It is important for any deployed system to be sensitive to threats but also maintain low false alarm rates, which is complicated by fluctuations in background radiation levels. This work seeks to show the efficacy of PNMF at identifying isotopes using gamma-ray spectra from small-volume CsI(Tl) detectors. Although this technique has been proven using larger NaI(Tl) detectors, covering a larger area with many smaller detectors is desirable within the UK as these detectors are of lower cost and are more readily available. In this work, spectra containing data for five radioisotopes are injected into a sparse background dataset, and are identified using reconstructions with PNMF components. A comparison is made between the results of successful identifications for 2 different source activities.