Research On Feature Extraction And Recognition Of Ship Radiated Noise

Underwater target classification and identification technique has important significance in the fields of national defense security and the development of marine resources.It is noteworthy that efficient ship identification method is very important for cruise law enforcement and the monitoring of infringing ship in sensitive sea areas.For ship target recognition,it is necessary to obtain radiated noise of ship target in the background of complex marine environment noise and extract feature parameters with class characteristics.The recognition based on a single feature is often not reliable enough,and simply distinguishing between two types does not satisfy the actual need for identification,so comprehensive analysis on multiple features in target identification is necessary.This paper discusses the suppression technology of the interference target signal in the acoustic field of view,studies the multiple feature extraction methods for ship targets,designs a passive classification and recognition framework for multi-class ship targets based on decision-level fusion,and realizes the effective classification of three types of ship targets.Significantly,the work is to obtain the high-quality original data of radiated noise at the beginning of ship identification,which needs to suppress the radiated noise of interference target in the acoustic field.In this paper,a vector acoustic shielding technique based on the virtual time reversal mirror theory is proposed,which could focus the target signal and shield the interference signal using a focusing-shielding weight filter in the spatial domain.Virtual time reversal mirror can realize acoustic focusing by multi-path channel.Meanwhile,vector signal processing can not only avoid port and starboard ambiguity,but also reduce side lobe interference.Simulation results verify the feasibility and effectiveness of the algorithm.Vector acoustic shielding beam-forming is a wideband processing algorithm.Compared with beam null-forming suppression technology,it does not need segment processing in different frequency bands and will not lose the phase information of the original signal.The superiority of the acoustic shielding algorithm is verified by comparing the power spectra of the two beam space output.In the study of ship target recognition,the propeller shaft frequency feature is stable,which has clear physical meaning.When extracting propeller shaft frequency from DEMON(Modulation of Envelope Modulation on Noise)spectrum,non-uniform modulation often causes line omission problem.Considering this situation,weighted fusion technology for DEMON spectrum is used by weighted processing of multiple sub-bands DEMON spectra.Based on weighted DEMON spectrum,the maximum common divisor algorithm is adopted to extract propeller shaft frequency of ship target.DEMON spectra weighted processing can not only make up the problem of missing line spectra,but also reduce the broadband background noise and enhance line spectra.In this paper,the propeller shaft frequency of multiple targets are successfully extracted and data processing result verifies the validity of the DEMON spectra weight fusion method.Aiming at the problem of low line SNR of DEMON spectrum,this paper proposes the interference suppression gate technique.The technique takes advantage of the fact that it is uncorrelated between sinusoidal signal and wideband noise,and the noise energy is concentrated at the correlation peak.The wideband interference is suppressed by zeroing the correlation peak value.The effectiveness of the algorithm is verified by simulation and real data processing.Interference suppression gate technique can be used for line spectrum extraction and detection.The type of ship can be reflected on the different continuous spectrum shape of DEMON spectrum.In this paper,through continuous spectrum smoothing and curve fitting,the polynomial coefficient of the fitted curve is taken as the ship target characteristic parameter.The five polynomial coefficients are regarded as the recognition feature parameter,which provide the foundation for the ship classification and recognition.In view of the shortcomings of the previous versions of Empirical Mode Decomposition algorithm,SN-EMD(Selective Noise-Empirical Mode Decomposition)is proposed in this paper.Through the simulation of deterministic signal decomposition,the performance of several algorithms is compared according to the performance evaluation index.The results show that SN-EMD algorithm has good anti-aliasing performance,better real-time performance than other noise auxiliary algorithms,and the residual error and reconstruction error are also relatively minimal.The center frequency of each mode component obtained by decomposing ship radiated noise does not obey the rule of dichotomy,and that the difference reflects the class characteristics of ship radiated noise exactly.In this paper,SN-EMD was carried out to decompose real radiated ship noise.Several mode feature parameters,including mean instantaneous frequency,center frequency,energy density,energy distribution and mode energy entropy were extracted for ship classification and identification.A decision-level fusion recognition mode is established based on shaft frequency,envelope spectrum,mode frequency and mode energy.On the basis of ship feature samples,BPA(Basic Probability Assignment)function based on the posterior probability and recognition rate of SVM(Support Vector Machine)classifier is constructed.According to DSm T(Desert-Smarandache Theory)evidence theory,BPA values are decision-level fused,and then the recognition result is output after judgment.In order to reduce data redundancy and computation,feature dimension is reduced by PCA(Principal Component Analysis).The separability between features of different ship targets is verified by intra-class and inter-class dispersion.The processing realizes screening and effective verification for input feature data.Finally,the noise data of three ship targets were tested,and the results showed that the right recognition rate of multi-feature decision-level fusion reached larger than 96%.Compared with recognition method on the basis of single feature,decision-level fusion recognition takes advantage of multiple information sources and improves the accuracy of target recognition significantly.