Machine Learning-Based Capability for Radioactive Threat Detection and Identification

Physical Sciences Inc. (PSI) proposes to develop advanced machine learning (ML) algorithms to perform detection, classification and identification of radioactive sources in near real time based on analysis of data generated by radiation sensors. The algorithms will be packaged as of an integrated Advanced Learning-Enabled Radioactive Threat Search (ALERTS) software suite compatible with new networked systems such as SIGMA. The algorithms will be trained to extract salient features from radiation sensor data during operational source encounters to generate source characterization output without the use of predefined source templates. The ML-based approach will provide a complementary analysis channel to enhance current state of the art detection and identification algorithms. The ALERTS algorithms will work in conjunction with PSI’s operationally deployed Poisson Clutter Split (PCS) algorithm to significantly improve the performance of current and future sensor systems used by US Armed Forces and Law Enforcement Agencies in the search for radioactive threats, especially Special Nuclear Materials (SNM). Under this effort, ALERTS will be customized for integration with SIGMA networked sensor system as well as integrated with PSI’s other commercial radiological/nuclear products such as Mobile Urban Radiation Search (MURS) and PERM-Mobile.