Neural network approach for inverting velocity dispersion; application to sediment and to sonar target characterization

A novel inversion method based on a neural network is presented. The parame ter used for the inversion is the velocity dispersion of surface acoustic w aves. The velocity law is computed using time-frequency representation (Wig ner-Ville), which is a highly relevant tool for multi-component dispersive signal analysis. The same inversion method has been applied to two physical cases: inversion of geoacoustical parameters of seabeds and characterizati on of sonar target properties. The method possesses high accuracy performan ce (error less than a few per cent) in both cases on either simulated or ex perimental data. The a priori information used for inversion is very poor a nd limited to the physical description of the direct problem.