Study On Condition Feature Enhancement And Extraction Of Ship Radiated Noise

Ship radiated noise is one of the most abundant information for underwater acoustic signals.After transmitting over long-distance underwater acoustic channel,it is inevitably interfered by ocean ambient noise and transient signals.How to extract ship radiated noise and its inherent signal characteristics from the weak underwater acoustic signals is one of the difficulties and hotspots for underwater acoustic signal processing.Research on condition feature enhancement and extraction of ship radiated noise plays an important role in performance improvement of underwater sonar recognition system.This dissertation will further improve the recognition effect of ship radiatied noise via the in-depth study of ship radiated noise condition feature enhancement and extraction technology.This thesis takes ship radiated noise as the research object,and focuses on the application of condition feature enhancement and extraction methods in ship radiated noise recognition with low SNRs.With the rapid development of ship vibration and noise reduction,noise elimination equipment and stealth technologies,the sound level of ship radiated noise has been reduced year by year,even it is completely submerged in ocean ambient noise at close range.The aboved problems make the performance of the traditional condition feature enhancement and extraction methods reduce significantly.Therefore,this thesis deeply studies the application of Resonance-based Sparsity Signal Decomposition(RSSD)and Manifold Learning(ML)theory in the condition feature enhancement and extraction of ship radiate noise.The main research contents of the dissertation include:(1)A condition feature enhancement and extraction method is proposed utilizing RSSD algorithm and Hilbert marginal spectrum(HMS)method,which is suitable for condition feature enhancement of ship radiated noise with low SNRs.First,the raw signal can be separated into high-resonance,low-resonance and residual components by RSSD algorithm,where high-resonance component is multiple oscillatory signal,low resonance component is transient signal without oscillation and residual component is noises.According to the low-frequency periodic oscillatory characteristic of ship radiated noise,its oscillatory signals can be well kept in high-resonance component.Meanwhile,the low-resonance and residual components are discarded to achieve the purpose of suppressing noise and transient interference.In addition,RSSD algorithm has the advantages of eliminating in-band noise and transient interference.Secondly,the feature vector of high-resonance component is extracted by HMS algorithm.Finally,support vector machine classifier(SVM)is adopted.Under different SNRs condition,the classification results verify that the proposed method has much better recognition performance than HMS method,and the average correct recognition rate is at least higher than the comparison method about 3.7%.(2)A resonance-based time-frequency manifold(RTFM)algorithm is proposed in a pioneering manner,which is suitable for condition feature enhancement of weak ship radiated noise with lower SNRs.The algorithm consists of the following four steps:first,RSSD is applized to the one-dimensional raw signal for extracting the high-resonance component which includes the vibrational information.Second,the phase space reconstruction(PSR)algorithm is used to convert the one-dimensional signal into the high-dimensional phase space,which preserves the non-stationarity and nonlinearity of the signal and also maps the noise to phase space.Third,Time-frequency distribution(TFD)method is used to construct the time-frequency images in the phase space;Finally,ML algorithm is employed to perform data dimensionality reduction on all acquired time-frequency signatures and obtain low-dimensional manifold information.Meanwhile,the obtained first and second low-dimensional manifold signatures are proportionally added,which has achieved better denoising effect.The proposed method does condition feature enhancement for both the one-dimensional raw signal and the two-dimensional time-frequency images of the phase space,and comprehensively considers the three characteristics of oscillation,non-stationary and nonlinear.In this thesis,experiments are carried out for two cases.One is to analyze two different types of ship radiated noise under the same SNR.The other is to test one ship radiated noise under different SNRs.The experimental results indicate that the performance of RTFM signature construction method is better than the time-frequency manifold(TFM)algorithm.(3)The pathbreaking work is converting feature extraction of one-dimensional signal into that of two-dimensional image using Gabor filter to extract features from two-dimensional RTFM,then obtain effective image texture feature vector.We use SVM classifier to recognize the extorted feature vector.The identification results verify the validity of the condition feature enhancement and extraction method based RTFM algorithm,and the correct recognition rate based RTFM method is higher than the TFM algorithm.In view of the above methods,the innovation work of this dissertation can be summarized into the following three aspects:(1)A condition feature enhancement and extraction method combining RSSD and HMS algorithm is proposed to deal with one-dimensional ship radiated noise with low SNRs.(2)RTFM algorithm is proposed in a pioneering manner.The proposed method comprehensively considers the three characteristics of oscillation,non-stationary and nonlinear,and does condition feature enhancement for both the one-dimensional raw signal and the two-dimensional time-frequency images of the phase space for the first time.This method is suitable for condition feature enhancement of weak ship radiated noise with lower SNRs.(3)The pathbreaking work is converting feature extraction of one-dimensional signal into that of two-dimensional image using Gabor filter to extract features from two-dimensional RTFM.The research in this thesis shows that the condition feature enhancement and extraction method is of great significance to improve the accurate perception and recognition performance of ship radiated noise.