Bayesian Network of Support Vector Machines for Robust Threat Identification

For any real world pattern classification system (detection, identification, and tracking), there is an inherent problem of optimizing the algorithm's performance for a wide range of scenarios. This is due to a large dynamic range of variables (solarloading, solar reflectance, atmospherics, background, target condition, countermeasures) which will impact the target's feature set. The variability of the feature set is very complex and will depend on time of day, local environmental conditions,previous weather conditions, signature reduction techniques to name a few. East West Enterprises Inc. proposes a novel approach for an adaptive fusion/discrimination algorithm architecture which integrates Bayesian Networks and Support Vector Machines(SVM). SVMs have been used for a variety of applications including detection, classification/recognition/identification, regression and density estimation. In many applications, the SVM has been shown to have superior performance over classicalstatistical and neural network algorithms. The structure of the SVMs incorporates the training sample size, number of features, and desired performance in order to give optimal generalization performance. A fusion architecture is proposed based onintegrating a network of SVM classifiers a Bayesian Network to allow for the adaptive processing/fusion of data from multiple sensors. High fidelity fusion algorithms. These will provide great benefit to military and commercial applications involvingtarget detection, identification, tarcking and discrimination. Potential benefits include MDA/GMD/THAAD, military reconnaisance, surveillance and site monitoring. Commercial applications include homeland defense, medical, and industrial inspection