Inversion of Shallow Water Sound Speed Profile Using a Learning-Based Matched-Field Algorithm

This paper proposes a method for sound speed profile (SSP) inversion in shallow water based on a gated recurrent unit (GFGRU). The K-singular value decomposition method is used to generate a dictionary for the SSP representation. The major tasks of the GFGRU trained on synthetic and real acoustic datasets are multilabel classification and finding basis functions representing the disturbance of the SSP from the dictionary. Next, a genetic algorithm is used to search the coefficients of the basis functions and obtain the inversion SSPs. The performance of the proposed inversion method was compared to the traditional matched-field inversion method, classification results, and final inversion results of the simulation acoustical dataset, driven by the measured SSPs in Shallow Water Experiment 2006, to verify the accuracy and robustness of the proposed learning-based matched-field method.