Performance of Neural Networks in Classifying Environmentally Distorted Transient Signals

Neutral networks have been showing great promise in several areas, one of which is the classification of underwater acoustic transients. The classification of low-frequency underwater acoustic transient signals using a neural network based system is investigated. The received acoustic transients are simulated using a time-domain parabolic equation model. The neural network is trained on three source signals and tested by classifying the same signals at 25 different receiver locations in a noise-free, range-dependent (upslope) environment. Overall classification performance is above 90%.