Neural Networks and Fuzzy Logic for Battlefield Sensor Fusion

Today's battlefield environment poses an increasing challenge due to more sophisticated and capable threats. What is required is to find a method that can detect, classify, and engage targets using multisensor data in a timely, efficient manner and which is robust to countermeasures, environmental effects, and sensor limitations and constrains. We propose a hybrid neural network/fuzzy logic/information processing approach for battlefield sensor fusion. Our approach is based on successful results using a hybrid approach. Our baseline Phase I effort will focus on development of a sensor database that includes sensor characteristics and capabilities versus mission requirements and identifies neural network/fuzzy logic implementation solutions. Additionally, we propose an optional feasibility demonstration by developing and evaluating a prototype system. Our hybrid approach to sensor fusion offers several advantages. Neural networks offer rapidly adaptable on-line solutions, capability to fuse multispectral channels and to detect and classify new threat classes, and implementation efficiency on emerging VLSI designs. Fuzzy logic complements neural networks in sensor calibration, interpreting network outputs, detecting and classifying threats, fusing decisions from parallel neural and rule base classifiers, and incorporating ambiguous information. Information processing algorithms perform sensor measurement pre-processing, data alignment, feature extraction, an data association.