【作者】 贾雪松；

【导师】 孙进才；

【作者基本信息】 西北工业大学 ， 水声工程， 2004， 硕士

【摘要】 水下目标被动识别技术一直是水声信号处理领域研究的热点。长期以来，人们采用各种方法提高水下目标的识别率。虽然识别性能不断改善，但根本性问题并没有解决。这是由于水下目标辐射信号和海洋环境的复杂性、多变性造成的。 对于水下目标辐射噪声的产生机理，可以发现它们与相关动力学系统的结构有关。非线性时间序列分析方法提供了在相空间中研究这些动力系统生成的信号的有效途径。依靠非线性时间序列分析方法提取重构动力系统的非线性或混沌特征参数，可以达到识别的目的。本文以非线性时间序列分析和混沌理论为基础，围绕舰船辐射噪声的特征提取这一主题，进行了如下工作： 1．研究了相空间延时重构的两个参数的估计问题。利用平均互信息法确定重构时延参数，相比于自相关函数法，这个方法更加准确。采用伪最邻近点法确定最小嵌入维数，这种方法可以准确地确定含噪信号中确定性成分的最小嵌入维数，避免嵌入维数过高导致相空间的混乱。 2．在主元分析（Principal components analysis，PCA）的基础上，研究了非线性局部投影滤波方法，该方法的原理是在相空间分段线性逼近动力系统并进行局部主元分析。与一般的主元分析相比，非线性局部投影滤波方法的性能明显提高，在较低信噪比下仍能较好地恢复原信号的波形和相空间轨迹。 3．研究了时间序列的混沌特征参数提取方法。包括关联维数算法中的时间相关、最大Lyapunov指数的稳健估计算法以及广义维数和时间序列熵h₂的估计问题。通过对舰船辐射噪声进行混沌特征分析，结果表明，舰船噪声不具有严格分形性，但具有正的最大Lyapunov指数存在，这说明舰船噪声信号中有非线性成分存在。通过对不同类别信号的特征提取表明，关联维数和广义维数能有效地代表不同类别的舰船特征，而h₂熵特征并不明显。更多还原

【Abstract】 The passive recognition technology of underwater targets is always a hot topic in the field of underwater signal processing. For a long time, scientists have been improving the recognition technology of underwater targets by variant methods. Although the performance of recognition has been improved, the key of the question has still remained unresolved. This is caused by the complexity of underwater target radiated noise and the variety condition of the sea.After studying the mechanism of the underwater targets radiated noise, we learned that the signals are related to the correspondent dynamics. While the nonlinear time series analysis methods offer the technologies to process the underlying dynamics of the signals in the phase space. The targets recognition can be reached by extracting the features of the reconstructed dynamics using nonlinear methods. In this paper, some work has been done to extract the features of the ship radiated noise signals on the basis of the nonlinear methods and chaos theory. The main contributions of the dissertation are as follows:1.Two parameters of delay reconstruction in phase space are explored. The optimum delay time is evaluated by the average mutual information method; this method is more precise than the autocorrelation method. The minimum embedding dimension can be determined by the false nearest neighbors method, and embedding dimension of determination component in noise condition can be determined precisely by this method which can avoid the phase space disturbance because of the higher embedding dimensions.2.Nonlinear local project noise reduction is studied on the basis of principal components analysis (PCA), the principle of this noise reduction is to approximate piecewise and linearly the dynamics and to process local principal components analysis. Contrast to the simple PCA, the filter performance of the nonlinear localproject noise reduction method has been improved greatly, it can recover the wave of the origin signal and trajectories of phase space even in low S/N ratio.3.Feature extraction of time series based on chaos theory is explored, which include the problem of temporal correlation in correlation dimension method, the robust method to evaluate the maximum Lyapunov exponents, the extraction ofgeneralised dimensions and the evaluation of h2 entropy of time series. It can belearned by analyzing the ship radiated noise signals using nonlinear methods that the ship radiated noise signals aren’t rigid fractal signals, but there are positive maximum Lyapunov exponents, this indicates that the ship radiated noise signals are nonlinear. The extraction test to the different class signals tell us that the correlation dimension and generalised dimensions may effectively involve the class information of differentships, otherwise the h2 entropy is difficult to classify the targets.更多还原