Background Adaptive Radiation Detection (BARD)

Various applications in nuclear non-proliferation, homeland security, and basic science require the use of statically deployed radiation detectors to monitor background radiation, detect and identify anomalous signatures. In the context of large city-scale deployments, it is being understood that domain awareness brought in by additional contextual sensors is required to provide enhanced detection performance and data interpretability. Such multi-modal sensor platforms, in conjunction with on-edge computing capabilities, allow the implementation of real-time algorithms for radiological anomaly detection, isotope identification, and source localization and tracking. This paper focuses on radiation data processing algorithms, whose performance is critical to reduce false alarms and misidentification over large detector networks. We present a robust, automated ML-based adaptive approach to in-situ ab-initio background learning and anomaly detection.