A neural network-based underwater acoustic application

A neural network-based detection and classification design has been developed for a real world underwater acoustic data set. The data includes a number of recorded examples of four different signals of interest plus environmental data. The design achieved better than a 90% probability of detection and correct classification on the signals while experiencing only one false alarm on the environmental data. A systems oriented development approach was employed, which included the use of a specially tailored computer aided development environment. Unlike the typical neural network application, the development made extensive use of expert knowledge