Research On Passive Location Of Underwater Sound Source Based On Artificial Neural Network

In recent years,passive positioning of underwater sound sources has been widely used in military and civil fields,and it is currently one of the popular directions of underwater robot technology.The previous passive location method of underwater sound source is mainly based on the environment and sound propagation model of the sound source to locate the target.Under normal circumstances,it is difficult to obtain real-time data of the underwater environment where the sound source is located,which affects the sound source location.Accuracy.This paper mainly receives sound source information through the underwater array,uses neural network to locate its position,uses the generalization ability of neural network to reduce dependence on the underwater environment,and learns the relationship between the array data and the sound source position from the sound source data,Complete the training of the sound source localization model,and then predict the target position of the sound source.This article mainly uses neural network classification and regression methods,as well as feature extraction of sound source data,and then uses neural network to estimate the target position.In this paper,the problem of passive localization of underwater sound sources is mapped to the classification and regression problem of artificial neural network to solve.Regarding the source localization problem as a neural network classification problem,research on feedforward neural network(FNN),support vector machine(SVM)network and probabilistic neural network(PNN)was carried out,and PNN was used for the first time in underwater sound source localization.Neural Networks.In the experiment,the particle swarm algorithm is used to optimize the parameters of the PNN,which ensures the reliability of parameter selection,and also verifies the influence of different snapshots on the sound source location.The experimental results show that the source localization is regarded as a neural network classification problem,and PNN has good accuracy for passive localization of sound sources.Secondly,the problem of source localization is further regarded as a regression problem of neural network.Mainly studied the generalized regression neural network(GRNN),RNN neural network and LSTM neural network.In this paper,the sound source localization problem is regarded as a supervised learning regression problem,and LSTM neural network is applied to the sound source localization problem.First,the sound source data is preprocessed as the input of LSTM,and then the LSTM is used to construct the sound source characteristics.Model,construct the regression model,join the early stopping method,improve the speed of network training,and finally use the trained LSTM network to predict the sound source location.The method is verified by simulation experiments and sea trial data.The experimental results show the effectiveness of LSTM for passive positioning of underwater sound sources.Finally,this paper makes further improvements to the sound source feature extraction.The eigenvalue decomposition is carried out through the covariance matrix of the multi-channel array element,and the main feature vector is extracted as the input of the neural network.At the same time,Bi-LSTM is introduced into the sound source localization,so that The algorithm has a stronger ability to locate dynamic sound sources.Combined with the above PNN neural network and LSTM neural network,it uses the extracted sound source features as input.The experimental results show the effectiveness of using eigenvalue decomposition on the sound source features,Which reduces the training time of the neural network,and at the same time verifies the feasibility of the method in passive localization of the sound source.