Underwater acoustic target recognition method based on genetic algorithm optimized BP neural network

Abstract

The invention provides an underwater acoustic target recognition method for optimizing a BP neural network based on a genetic algorithm. The method comprises the steps: extracting MFCC underwater acoustic features of an underwater acoustic signal, and enabling the underwater acoustic features to serve as a sample of a BP neural network classifier; randomly initializing a population of a genetic algorithm; wherein the individuals in the population comprise weights and thresholds of the BP neural network needing to be optimized, the weights and the thresholds serve as target optimization functions of the genetic algorithm, the target optimization functions are substituted into the BP neural network to obtain simulation errors, the simulation error of each individual in the population is optimized, and the optimal initial weight and threshold of the BP neural network are obtained; and finally, training and identifying the underwater acoustic features by using the BP neural network optimized by the genetic algorithm to obtain the accuracy of the classification algorithm. The method uses the genetic algorithm to identify and classify the characteristics of the underwater acoustic target, uses the BP neural network optimized by the genetic algorithm to train and predict the underwater acoustic target to obtain a higher recognition rate, and the recognition performance is more stable.

Description

technical field [0001] The invention relates to the field of underwater acoustics, in particular to an underwater acoustic target recognition method. Background technique [0002] In the ocean, all kinds of information are transmitted and obtained mainly by sound waves. In underwater acoustic target processing, the recognition and classification of underwater acoustic targets has always been one of the important research directions. [0003] Underwater acoustic target recognition technology mainly includes feature extraction method and classifier design. The task of feature extraction is to select effective, stable and reliable features that represent the identity of the target. The actual underwater target signal not only has a complex sound mechanism, but also has various components. The characteristic analysis method of underwater acoustic target has experienced the development from simple time domain and frequency domain analysis to time-frequency joint domain analysi...

AI Technical Summary

Problems solved by technology

[0005] However, in the actual application of the BP neural network, the selection of the network topology and the determination of the initial weight lack a theoretical basis, and its training mainly relies on the BP algorithm, which has the problems of slow convergence, easy to fall into local extremum and "over-learning".

Images