Anomaly Match Bayesian Aggregation (AM-BA) for Efficient Radiation Source Search

Abstract

Mobile radiation detector systems are used to detect the presence of sources of dangerous radiation. Efficiently searching over large numbers of source hypotheses is a major computation challenge since a source scene may have many possible source location hypotheses. Additionally, sources can have many other characteristic parameters such as intensity and type. Bayesian Aggregation (BA) allows simultaneous detection of radiation sources and inference of properties. In this study, we develop Anomaly-Match BA (AM-BA), a BA variant that saves computation in searching parameter spaces of source hypotheses.

Anomaly detectors and match filters are background estimators commonly used on radiation observations to separate estimated source components from typical background fluctuation. Anomaly detectors assume no knowledge of the source, simply finding anomalies to typical background fluctuation. Match filters, in contrast, use knowledge of a source template or design in estimating components. Both types of estimators are useful in BA for estimating and assembling distributions of the Signal-to-Noise Ratio (SNR) of measurements. In AM-BA, anomaly filters are first used to prune parts of the source location hypothesis space that are not likely to contain the source. Match filtering parameter optimization for other source parameters (e.g. intensity and type) is subsequently run on the smaller space of source hypotheses. The developed algorithm can save significant amounts of computation in searching the space of source parameter hypotheses without loss of threat detection capability.