Discrimination method of convergence zones in deep sea based on machine learning

Abstract

When the underwater acoustic targets are in different convergence zones, the characteristics of the acoustic signals received by the hydrophone have a strong similarity. The traditional Matched Field Processing (MFP) localization method is difficult to discriminate the convergence zone that the underwater target exists. Based on the different characteristics of the spatial-frequency structure of the sound field in the convergence zone, this paper proposes a discrimination method by using the classical convolutional neural network (CNN) model. Acoustic propagation model is used to generate simulated sound field, which replace the experimental data to train the model. Compared with the MFP, the machine learning based has certain tolerance to the uncertainty of environmental parameters. Computer simulation data and experimental data verify the effectiveness of the algorithm.