An underwater target recognition method based on the fusion of GRU and one-dimensional CNN neural network

Abstract

The invention discloses an underwater target recognition method based on the fusion of GRU and one-dimensional CNN neural network, belonging to the field of underwater acoustic target recognition. Aiming at the problem of underwater acoustic target recognition, an underwater target recognition method based on the fusion of GRU and one-dimensional CNN neural network is proposed, and the problem that the traditional neural network cannot extract the temporal characteristics of underwater acoustic signals is solved by using the GRU-based cyclic neural network structure. , while using a one-dimensional CNN convolutional neural network structure to extract the time-domain waveform features of the underwater acoustic signal. The eigenvectors extracted by the fusion of GRU and one-dimensional CNN neural network structure enrich the eigenvalue information of the input classifier. The auxiliary information input that affects the recognition accuracy is expanded, including distance, hydrophone depth, channel depth and other information, and the dropout layer and batch normalization layer are added to avoid the overfitting problem and improve the recognition accuracy of underwater acoustic targets.

Description

technical field [0001] The technical field of the present invention is the field of underwater target recognition, specifically an underwater target recognition based on the fusion of a gated recurrent unit (Gated Recurrent Unit, GRU) neural network and a one-dimensional convolutional neural network (Convolutional Neural Networks, CNN) method. Background technique [0002] In the modern underwater acoustic system, the underwater acoustic signal processing technology has a huge content and wide coverage, which is the fundamental basis of underwater acoustic countermeasures. Many fields such as target detection, target feature extraction, and target feature identification belong to the category of underwater acoustic signal processing. Due to the particularity of the underwater environment, the application of underwater acoustic signals for underwater target detection is currently the most effective means. In underwater defense and military activities, the ability to effectiv...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to propose a kind of underwater target recognition method based on the fusion of GRU and one-dimensional CNN neural network for the underwater target recognition problem that requires increasing recognition accuracy; The problem that the network cannot extract time features; use the one-dimensional CNN structure to extract the time-domain waveform features of the underwater acoustic signal, increase the characteristic information of the underwater acoustic signal; increase the input of auxiliary information that affects the recognition accuracy, improve the recognition accuracy; and add Dropout layer and batch normalization layer to avoid overfitting problem

Images