Model of Hydroacoustic Signal Synthesis Using Neural Networks

Abstract

Underwater acoustics (also called hydroacoustics), which is associated with the study of the patterns of propagation of sound waves in water, is a driving force in the research and development of systems of hydroacoustic technologies and means of communication, monitoring and detection of surface and underwater objects of biological or artificial origin, study of marine resources and environments, noise measurement, etc. This kind of research requires the analysis of huge amounts of data, revealing non-obvious patterns and creating models for the mathematical description of physical phenomena, such as sound propagation in a medium with random characteristics and radiation from different sources, as well as radiation from sources with different apertures or sound scattering, etc. That is why, in order to create the latest technologies in this area, it is necessary to solve complex specialized problems of a fundamental and applied nature using machine learning algorithms and artificial intelligence. Neural networks are nonlinear systems that allow you to effectively classify data compared to mathematical and statistical methods, which are currently quite widely used. In this paper, the authors propose to use a pre-trained neural network for the analysis and classification of hydroacoustic signals. This procedure for distinguishing acoustic signals has a number of advantages, in particular, individual objects are divided into groups based on information about one or more characteristics inherent in these objects, and on the basis of a training sample of pre-labeled objects. Thus, the proposed model of signal synthesis using neural networks is characterized by increased informativeness of the characteristics of the propagation of hydroacoustic signals, which will have prospects in further practical implementation.