Low-Fidelity Spectral Analysis Utilizing a Binomial Discriminator for Weak-Source Detection Decisions

Abstract

The identification of radiological sources by analysis of a gamma spectrum usually relies on the location of the set of radionuclide-specific electron energies corresponding to the incident photons interacting by photoelectric absorption in the detection medium. The challenge in low-level detection applications is the identification of these "photopeaks" above the background counts registered in the detector from the natural radiation environment and system noise. For source detection decisions, regions of the gamma spectrum other than at the photopeak energies may provide additional information about the presence of a source and allow for a higher rate of correct identification of a weak source. A statistical algorithm utilizing low-fidelity spectral data partitioned into three distinct regions and employing a binomial discriminator was tested in a laboratory setting against the traditional approach of source identification by exceeding a decision threshold within the photopeak region of interest. For an unshielded Cs source with no significant scatter between the source and the detector, the traditional peak identification method performs as well or better than most algorithm settings for various source strengths. However, an algorithm which also includes information in the energy range of Compton scattered photons provides improved detection capabilities for shielded weak sources. Such algorithms, including higher-fidelity developments, could be deployed to improve current tools for the search for orphan radiological sources and in the characterization of low-level environmental contamination.