Source Position Sensitivity of a Coaxial HPGe Detector via Machine Learning

Spatial localization of gamma sources is often attained by using absorbing collimators or specifically-constructed detectors. If an estimate of the location of the gamma source could be determined using a standard portable Germanium (HPGe) detector without additional collimation, it would improve the efficiency of nuclear safety inspections in detecting illicit gamma-emitting material. Here, we propose to utilize machine-learning-based approaches to derive the direction of the gamma photon from the shape of the pulses produced by the HPGe detector. This shape is reflective of the location at which the photon deposits energy in the coaxial detection volume, thus providing a relationship with the direction of entry of the gamma photon. The shape-direction relationship is utilized by training a self-organizing map to develop patterns specific to the location of the gamma source. We use these maps to train another simple network to map these shape patterns to the direction of the gamma source and show that coaxial HPGe detectors are capable of estimating the direction of a gamma source.