Speedy UAV Swarms Detection, Identification, and Tracking using Deep Learning-Based Fusion Methodology for Radar and Infrared Imagers

Swarm UAS operations by sophisticated state and non-state actors are a cause for concern to the U.S. Naval missions.   The threat potential is not fully known. Some military theorists assert that “swarming” could be the Third Offset Strategy in military thinking and the basis for possible wide-scale impacts on military operations and tactics. Swarming is “a tactic that employs multiple small UAVs to overwhelm targets by using mass and attrition in combination with decentralized maneuvers or combined fires from multiple directions.” Swarms may be used to disrupt ship deck or aircraft flight operations, clutter the tactical picture, confuse situational awareness, precipitate inefficient or pre-mature expenditure of defensive weapons, conduct reconnaissance and intelligence gathering, obscure or mask high-threat units, overwhelm defenses, or themselves deliver micro-munitions including chemical or biological using coordinated ambush, circling, or multi-axis attacks. Swarms can be trained to exhibit intricate unpredictable biologic-like behaviors. Detection, identification, and tracking of UAS swarms is challenging because of deformability, adaptability, low small UAV (sUAV) Signal to Noise Ratio (SNR), large numbers of units, and elements of tactics such as ambush, deception, and remote targeting. Irregular grouping and ungrouping dynamics of complex coordinated UAS Tactics, Techniques, and Procedures (TTP) complicate the swarm picture, as the arrangement can change fluidly. The overarching objective of this SBIR topic is to develop sensor Data Fusion (DF) methodology for a phased array radar system and MWIR/LWIR infrared cameras to combine data from different and orthogonal modalities in support of an air defense system.  A specialized DF function is the identification of hard to detect and recognize objects for which Deep Learning (DL) has in recent years shown to be a good solution.  Thus, the target fusion process design and prototypes comprise a hybrid DF + DL architecture that is innovative in several ways:Innovation # 1: Radar Intrapulse Super Resolution for unresolved targets using DL trained on synthetically superposed sUAV inverse backscattering and sUAV Doppler signaturesInnovation # 2: Group Tracking using a novel spatial frequency technique using DL trained on volumetric Fourier transforms of swarm formations, an approach that avoids problems inherent in using multi-target association for swarmsInnovation # 3: Multi-swarm mission association and behavior inference using DL trained on Swarm TTPRadar and IR sensing are made complimentary and are used to form swarm tracks without ever having to track individual sUAVs. Associating and tracking at the swarm level allows low visible, dim, and weakly resolved image detections to be merged into the swarm track.  This blend of sensors and swarm signature-based DL and DF has never been tried before but is well-founded and solves many of the long-standing issues with swarm tracking.