A Neural Network Model-Based Inversion Method for Stratified Seafloor Geoacoustic Parameters in Shallow Water

Abstract

At this stage, in the inversion research of shallow sea geoacoustic parameters, the traditional inversion method focuses on calculating the parameter inversion under the default seafloor layering, and does not clarify the seafloor layering. Therefore, when dealing with the experimental data under the unknown seabed stratification, the geoacoustic parameters that conform to the specific seabed stratification cannot be accurately given. Aiming at such problems, this paper proposes a layered shallow sea geoacoustic parameters inversion method based on a layered physical model of the seafloor and a neural network algorithm model. First, a subsea physical model with a layered structure is selected through simulation, and the fast field sound field method (FFM) is used to calculate the sound field, then the theoretical prediction value of the shallow sea acoustic pressure field under different layered conditions is obtained. Second, the pattern recognition neural network algorithm is established to analyze the characteristics of the predicted sound pressure field under different seabed layers. Finally, according to the layered results, the sound pressure field is substituted into the seabed parameters inversion model of the corresponding layers, and the seabed parameters inversion is performed to obtain the inversion results. Experiments results show that the number of model iterations in the neural network algorithm is less and the efficiency is higher. The problem of seabed stratification is effectively solved through the characteristic analysis of the sound pressure field, which greatly improves the convenience of the inversion work.