Geoacoustic model inversion with artificial neural networks

Abstract

An ocean parameter estimation methodology is presented which involves neural networks. Both multi-layered perceptron networks and radial basis function networks were trained to estimate ocean bottom parameters from a received acoustic signal. The network's design algorithms are presented and their relative merits discussed. The pre-processing of the data is described in detail. A comparison of the relative accuracies of the two networks for simulated data is presented. The inversion of actual data from the TRIAL SABLE experiment was performed and the parameter estimates are given.