Shallow Water Environmental Parameter Inversion Method Based on Multipath Fusion Network

Aiming at the problem that the traditional matching field inversion method is computationally complex and inaccurate in the inversion of the underwater test environment, this paper proposes a multi-path fusion network (MPF network) based on the residual network for shallow water environment inversion. The method fuses residual features in different paths by introducing different upsampling methods between residual blocks. The method can fuse the input feature data more efficiently. Experiments show that, compared with the normal wave model and parabolic equation model in the traditional matching field inversion method, the method proposed in this paper has 70%+ and 50%+ improvements in the inversion of sound velocity profile and temperature profile, respectively.