Robust Multisource Automated Data Fusion and Target Identification

Reliable Non-Cooperative Target Recognition (NCTR) techniques exist for many different sensors. The effectiveness of each technique is limited due to restrictions placed on the target characteristics. We propose algorithms that fuse target ID informationfrom different NCTR sources in order to improve the overall NCTR performance regardless of the target characteristics. During the Phase I effort, we developed and tested several fusion approaches on simulated data. Without utilizing informationregarding the quality of the individual NCTR technique, the results indicated that the superiority of the FUSION technique (e.g., Bayes) is prominent for data of good quality, whereas for data of poor quality, training or pre-processing (e.g., confusionmatrix, correlation) is more important. Incorporating training, estimation, and information regarding the quality of the individual NCTR technique renders the Correlation and Confusion Matrix methods the best, followed by the Dempster's Rule Fuser. Theprimary focus of the Phase II will be to design and develop a prototype software system consisting of robust multisource data fusion and NCTR algorithms. The emphasis will be on further prototype development and demonstration of the data fusiontechniques implemented in Phase I. Additional fusion approaches including the Fuzzy Modified Dempster's approach will also be developed under this effort. This Phase II development will be based on testing against REAL DATA with target identificationinformation from multiple NCTR techniques. The project team includes Dr. Ronald Mahler of Lockheed Martin, who will provide both technical and commercialization support.