Sensor fusion and classification of acoustic signals using Bayesian networks

In order to maximize the probability of detection and classification of a sonar contact, particularly in an adverse acoustic environment such as shallow water, sensor fusion is critical because (1) one sensor type may not be sufficient to detect all target classes, and (2) the statistical properties of the sensor fusion algorithm enhance the classification performance beyond that achievable with the best individual sensor. We consider the problem of multiple sensor correlation and fusion using Bayesian networks in a hierarchical scheme, considering first two different active sonar waveforms, then combining active and passive sonar, and finally information from other sensors and sources.