Detection and characterization of potentially harmful radiation sources is one of common problems encountered in the context of homeland security as well as in safeguarding of radioactive isotopes used in industrial or medical practices. Any improper or insecure storage of radioactive material may cause a substantial harm to humans and to the environment. The task of detecting relatively weak, potentially shielded sources, is particularly difficult when facing variability of patterns of background radiation. This is a common circumstance in cluttered urban environments. It is also in these environments where any improperly stored radioactive material may inflict the most extensive harm. In this thesis, we explore two algorithms that are useful in different situations. The first, the G-P detection method, works well with large sensors that can fit in vehicles. Its novelty is in making Poisson assumptions about the photon counts observed in gamma-ray spectra, while making Gaussian assumptions about their rates. The second, List Mode Regression, is tailored to small portable sensors which produce spectra with low photon counts. Both methods outperform current state of the art algorithms in their respective usage scenarios.

37 pages

Thesis Committee:
Artur Dubrawski (Chair)
Srinivas Narasimhan

Frank Pfenning, Head, Computer Science Department
Andrew W. Moore, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science