Acoustic Features Adaptive Extraction And Application Of Underwater Dynamic Targets

The acoustic characteristics of underwater dynamic targets including unmanned underwater vehicles,are the basis of underwater acoustic signal processing.Thereinto,the line spectrum features have the characteristics of strong energy,stability,and long propagation distance.So they become an important basis for detecting weak or distant targets under unmanned autonomous conditions and crucial for achieving target detection and tracking.Therefore,how to extract accurate line spectrum features has always been a research hotspot in the field of underwater acoustic signal processing.However,due to the complexity and diversity of the underwater environment,there are still many problems in line spectrum extraction.Firstly,there are a large number of sound sources and interferences in the marine environment,resulting in weak line spectrum features and low signal-to-noise ratio in underwater acoustic signals.Secondly,there are many target line spectra,which makes a poor real-time performance of line spectrum extraction methods.At the same time,the relative motion between moving targets and passive sonar produces a Doppler effect,which causes frequency offset in the line spectra and increases the difficulty of line spectrum extraction.Finally,low signalto-noise ratio and incomplete collection of the target’s complete motion cycle will affect the completeness of the line spectrum features.To address the above problems,this study conducted the following research:1.In response to the problem of low signal-to-noise ratio in underwater acoustic signals,a preprocessing method based on superimposed variational mode decomposition residual is proposed on the basis of the original variational mode decomposition-based signal preprocessing.By including the residuals obtained from the original signal decomposition in the process of signal reconstruction to obtain preprocessed signals,this method can achieve higher signal-to-noise ratio and smaller root-mean-square error signals,while better preserving the line spectrum features.2.As for the problem of poor real-time performance and line spectrum frequency offset,a line spectrum center frequency extraction method based on narrowband processing is proposed.On the basis of the preprocessed signal,the line spectrum frequency band is determined based on the energy criterion,and the center frequency of the line spectrum can be estimated by correcting the frequency of the strongest power within the frequency band.Then,the obtained frequency band are narrowed by using the Doppler frequency shift calculation formula.Combining time-frequency analysis with spectrum refinement,the line spectrum instantaneous frequency can be obtained,which can reflect the dynamic characteristics of the line spectrum.3.In response to the problem of incomplete line spectrum features,considering the nonlinearity,non-stationarity,and autocorrelation of line spectrum features,time series prediction methods based on LSTM and logistic regression are introduced to optimize the extracted line spectra.By using the accurately extracted line spectrum instantaneous frequency as the training set and the predicted results as the line spectrum instantaneous frequency of the subsequent moment,more accurate line spectrum features can be obtained.This article divides the line spectrum extraction problem into the above three issues according to the processing data flow sequence and provides solutions for each of them.The open-source dataset processing results show that compared with other methods,the proposed preprocessing method can significantly improve the signal-to-noise ratio,the line spectrum extraction method can quickly and adaptively obtain all line spectrum features,and the time series prediction method based on LSTM can optimize the line spectrum features more accurately.On this basis,the line spectrum extraction results of the measured data and the obtained target speed are compared and analyzed,verifying the practical application value of the proposed method.