Discriminating between dismounts, fauna, vehicle, blowing vegetation, and other moving objects detectable by Ground Moving Target Indicator (GMTI) radar is of great importance for many surveillance and reconnaissance tasks. Current state-of-the-art discrimination algorithms usually involve range-Doppler signature methods involving long sensor dwell durations, but due to practicality issues, methods not dependent on high sensitivity and long dwell durations are desirable. In particular, behavioral patterns visible in long observation intervals may be exploited to discriminate between target classes. Toyon Research Corporation proposes a dual-layer approach to this problem. A training-based method using a classifier performing supervised learning forms a large component of the lower-level classification in regard to variation in acceleration, signal-to-noise ratio, and other such general parameters. Output from this classifier forms part of the input to a model-based classification method implemented by a particle filter as the upper level, discriminating based on such criteria as starting position, no-go regions, and other such specific parameters.
Keywords: Gmti, Radar, Discrimination, Target, Dismount, Machine Learning, Particle Filter
