Comparison of multilayer perceptrons and maximum likelihood processors in the context of seafloor parameter estimation

The objective of this work is acoustic seafloor characterization based on the statistical properties of bottom reverberation. The seafloor roughness parameters are related to backscattering strength through the Helmoltz-Kirchhoff scattering model. These parameters are estimated through the application of multilayer perceptrons and maximum likelihood (ML) estimators to synthetic backscatter representing seafloor with different morphology. The multilayer perceptron is presented with synthetic sequences and is instructed to yield the parameters of interest in the output. ML estimation relies on the construction of a rectangular grid that maps scattering strength vectors to the two parameters under estimation through the Helmholtz-Kirchhoff model. The parameter estimates are the coordinates of the grid element at which the likelihood function-calculated for sets of simulated observations-obtains a maximum. The performance of both estimation processors is determined by the mean and variance of the obtained parameter estimates