Convolution neural network ranging method in the deep-sea direct-arrival zone

The deep Convolution Neural Network(CNN) is used to learn the characteristics of Vertical Linear Array(VLA) sound field domain(CNN Field) and VLA beam domain(CNN-CBF) respectively to estimate the source distance in the deep-sea direct-arrival zone.Firstly,the simulated sound field data is preprocessed,and then the sound field-domain and beam-domain data are used as training sets to train the CNN model.Finally,the test data are input into the trained model to estimate the source range.The simulation results show that the ranging results of the CNN-Field method are quite different under the test set of different seabed parameters,and the difference of the CNN-CBF method is small,and the estimation accuracy of CNN-CBF in a 16-element VLA with element spacing of 10 m can reach 97%when the signal-to-noise ratio is greater than 0 dB.The sea trial data processing results show that the ranging accuracy of CNN-CBF in the deep-sea direct-arrival zone is higher than that of CNN field,and the average accuracy rate can reach 93.16%within 10 km.