Machine Learning

DESCRIPTION:
Under a previous MDA SBIR project (Contract No. W9113M-07-C-0106), Virtual EM explored the application of Support Vector Machines (SVMs) as a novel optimization technique in solving real-time missile target identification problems. A feasibility study was completed based on a simplified model of real-world scenario in order to demonstrate the capability, efficiency and robustness of SVMs. Two types of SVM-based classifiers were studied using simulated radar data generated by full-wave electromagnetic modeling of representative missile and decoy geometries using VirAntenn™ software. All missile and decoy parameters including projectile were based on publicly available information. Various Signal-to-Noise Ratios (SNRs) and data sampling rates were considered. SVMs demonstrated robust performance in picking missile targets among decoys. Target and decoy were correctly identified 80% of the time for Signal to Noise Ratio (SNR) values above 0dB, and the accuracy increased monotonically for higher SNR values.

SPONSOR: Missile Defense Agency

SBIR TOPIC #: MDA06-033 (Phase I)

COLLABORATORS: University of New Mexico