Multi-Phenomenology Discrimination for Feature Aided Data Fusion

In this Phase II effort we will produce a multi-sensor discrimination system using a Manifold Learning (MFL) algorithm originally developed under a prior effort.Whereas other discrimination systems utilize features defined by subject matter experts (SMEs), our MFL algorithm learns important characteristics of a target through training with historical data.In prior work, we have applied our MFL algorithm to single sensor returns of radar and Electro-Optical Infrared (EO/IR) data to perform discrimination.In this work we will utilize the manifold structure within a high dimensional vector space to link the returns from multiple sensors in a novel and effective manner.This manifold-mapping capability will allow us to perform joint discrimination and correlation between sensors that exploit different wavelength regimes and phenomenology.This proposed capability is a game changing feature of our algorithm, in that this system could perform feature aided tracking without pre-defining what features are to be used.The benefits of our algorithm would be twofold.First the algorithm will still prove useful in situations where SME derived features are denied or unavailable.Second, and possibly more importantly, the algorithm could be used on current or new sensors regardless of the phenomenology exploited.Approved for Public Release 14-MDA-7739 (18 March 14).