Radioactive source localization using a data driven MVA method

Abstract

We present a small form factor (0.5 cm3) static CdZnTe (CZT) sensor network consisted of a number of Non Directional Detectors (NDD) capable to localize a stationary radiation source in 3D. The localization is performed with fusion algorithms based on AI techniques. The algorithms are based on Multilayer Perseptron Neural Network (MLP) and Gradient Boosted Decision Trees (BDTG). A data driven method based on experimental data was used to produce the events for both training and testing of the algorithms. The localization efficiency of the algorithms was verified with a different set of experimental data taken in our laboratory using a 137Cs source of 180 μCi. A localization resolution of a bare radioactive source of the order of 10 cm to 20 cm has been archived in 3D after at least of 40 sec exposure time within a monitored volume of 5 m × 2.8 m × 2 m. The localization of slightly shielded sources in 3D can be also achieved but with a hardly worse resolution.