Research On Nonlinear Dynamical Feature Extraction Method Of Ship-radiated Noise

Ship-radiated noise is a signal emitted by a ship during navigation,and nonlinear dynamics feature extraction for ship-radiated noise is beneficial to the detection,identification and tracking of ship targets,which has important research significance.However,the ship-radiated noise has three characteristics of non-Gaussian,non-linear and non-smooth,and the traditional nonlinear dynamics features have the problems of incomplete information characterization,poor differentiability and single feature.To address the above problems,this paper carries out the research of nonlinear dynamics feature extraction method based on dispersion entropy and Lempel-Ziv complexity(LZC),which mainly includes:1.To address the problems of incomplete information representation and poor differentiability of dispersion entropy,a new entropy-based nonlinear dynamics index named simplified coded dispersion entropy is proposed.The algorithm introduces quadratic partitioning on the basis of dispersion entropy to improve the separability of dispersion entropy on the one hand;on the other hand,the quadratic partitioning criterion is simplified to reduce the computational consumption of the algorithm.Simulation and actual test results show that:compared with dispersion entropy,permutation entropy and coded permutation entropy,the simplified coded dispersion entropy can more accurately reflect the dynamic changes of the simulated signal,changes in the degree of confusion and changes in frequency composition components;for the actual measured ship-radiated noise,it also has a higher classification recognition rate.2.To address the problems of effective information loss and poor separability of LZC,a new LZC-based nonlinear dynamics indicator named dispersion entropy-based LZC is proposed.The algorithm introduces normal cumulative distribution function on the basis of LZC to reduce information loss on the one hand;on the other hand,it replaces elements with dispersion pattern labels to further improves the separability of the features.Simulation and actual test results show that:compared with LZC,dispersion LZC and permutation entropy-based LZC,dispersion entropy-based LZC can reflect the dynamic changes of the simulated signal and the changes of the chaos degree more accurately,and has a stronger resistance to noise interference;for the actual measured ship-radiated noise,it also has a higher classification recognition rate.3.A hybrid multi-scale feature extraction method is proposed for the problem of single feature of simplified coded dispersion entropy and dispersion entropy-based LZC.The method introduces coarse-grained processing on the basis of simplified coded dispersion entropy and dispersion entropy-based LZC to increase the number of features on the one hand;on the other hand,the features with maximum correlation and minimum redundancy are selected to improve the computational efficiency of the method.The results of the actual experiments show that the proposed method has higher classification recognition rate in ship-radiated noise compared with dispersion entropy,permutation entropy,coded permutation entropy,LZC,dispersion LZC and permutation entropy-based LZC.