Passive sonar recognition and analysis using hybrid neural networks

The detection, classification, and recognition of underwater acoustic features have always been of the highest importance for scientific, fisheries, and defense interests. Recent efforts in improved passive sonar techniques have only emphasized this interest. In this paper, the authors describe the use of novel, hybrid neural approaches using both unsupervised and supervised network topologies. Results are presented which demonstrate the ability of the network to classify biological, man made, and geological sources. Also included are capabilities of the networks to attack the more difficult problems of identifying the complex vocalizations of several fish and marine mammalian species. Basic structure, processor requirements, training and operational methodologies are described as well as application to autonomous observation and vehicle platforms.