Neural network data fusion concepts and application

A neural network data fusion decision system and the accompanying experimental results obtained from joint sensor data are presented. The decisions include detection and correct classification of space object maneuvers simultaneously observed by two radars of different aspect, frequency, and resolution. The system consists of a statistically-based adaptive preprocessor for each sensor, followed by a highly parallel neural network for associating the preprocessor outputs with the appropriate decisions. The preprocessing approach, supported by a signal decomposition theorem, recursively models the detrended sensor data as an autoregressive process of sufficiently high order. This approach also accommodates nonstationary data by incorporating an information-theoretic transition detector which identifies the segments of near-stationary data. Together, feature vectors are produced over near-stationary segments of data which are scale invariant, translation invariant, and normalized and represent sufficient statistics. Subsequently, the feature vectors arising from the sensor preprocessors are collectively associated with the correct output decision. The association is conducted by a multilayer perceptron neural network associative memory employing a modified learning algorithm which converges at a rate comparable to that of conventional algorithms, yet requires less computation