Research On Ship Identification Based On Passive Underwater Acoustic Signals

The ocean area accounts for about 71% of the earth ’s surface area,and the ocean contains abundant resources.The exploitation and utilization of Marine energy and resources by human beings is gradually expanding,which makes the importance of ocean in economy and military more prominent.Ships are the main platforms for the development and utilization of the ocean by human beings.When ships are sailing,they will emit noise and radiate to the surrounding area,and of all the known forms of energy,sound travels best in water.The detection and classification of ship radiated noise is an important research topic in the field of underwater acoustic engineering,and it is also the key to the intelligent processing of underwater acoustics,which is particularly important in the field of national defense and military affairs.At present,the detection of underwater acoustic targets is mainly divided into active sonar detection and passive sonar detection.Active sonar detection has poor concealment and a short detection distance.Therefore,this paper mainly studies the underwater acoustic recognition based on passive sonar detection,and the main research technology is carried out from feature extraction and classification recognition.Firstly,this paper studies the common methods of feature extraction and classification recognition.Feature extraction methods include wavelet packet analysis and Hilbert-Huang transform(HHT),which are commonly used in processing nonstationary signals,and Mel frequency cepstrum coefficients(MFCC)which is widely used in speech processing.And then the principle and feature extraction process of the three methods are described in detail.Accordingly,the gaussian mixture model(GMM)is selected as the classifier,and the principle and training method of GMM are introduced.Secondly,the method of deep learning is applied to the recognition of underwater acoustic targets.Through the research of deep neural network,the convolutional neural network(CNN),which is popular in the field of computer vision,and LSTM,which performs well in the processing of long time series data,are applied to the recognition of underwater acoustic targets.In the research of CNN model,a CNN network model with3 convolutional layers is designed,and the input of the model is the sound spectrum,Mel spectrum and wavelet spectrum.In the study of the LSTM model,the network models of LSTM structure and the bidirectional LSTM(BLSTM)structure were designed respectively,and MFCC is used as the model’s time input for identification.In addition,because the recognition effect of BLSTM model is better than that of LSTM model,this paper also designs a network model based on the CNN-BLSTM structure,which combines the advantages of CNN in extracting local spatial features with the advantages of LSTM in processing time-series features.CNN-BLSTM model uses the same input as the CNN model for recognition.Finally,based on the above models,the recognition rate of each model with different signal-to-noise ratio is studied.Through comparison,it is found that with the underwater acoustic data used in this paper,the recognition rate of each model is relatively high without noise,with the average recognition rate above 90%,and the recognition rate of the CNN-BLSTM model using the wavelet spectrum as input is the highest,reaching99.6%.When the underwater acoustic signal-to-noise ratio decreases,the overall recognition result of the deep neural network model is better than that of the GMM model.