【作者】 王洪超；

【导师】 李亚安；

【作者基本信息】 西北工业大学 ， 水声工程， 2007， 硕士

【摘要】 海洋环境的复杂性以及舰船隐身降噪技术的不断发展，使得水声信号的检测变得愈加困难，传统的检测方法显现出了一定的局限性。近年来，随着混沌理论研究的不断深入，人们将混沌理论引入到水声信号的分析中来。同时，结合水声信号的特点，研究了水声信号产生混沌的机理，证明了水声信号的混沌特性。本文在此研究的基础上，对水声信号的混沌特征参数提取，混沌信号的降噪处理以及混沌信号的检测与识别等进行了进一步研究，本文主要研究内容如下： 1．针对水声信号的非线性特点，研究了基于局部投影滤波理论的水声信号的降噪算法。重点对局部投影滤波算法中邻域半径参数的选择进行了讨论，提出了基于递归分析和利用噪声强度计算邻域半径的两种方法。采用这两种方法，分别计算含有不同噪声强度的Logistic、Lorenz以及水声信号的邻域半径，进行局部投影降噪处理。结果表明利用噪声强度计算邻域半径的方法具有较好的降噪效果。 2．研究了水声信号混沌特征参数的计算，通过比较降噪前后水声信号混沌特征参数的差异，分析了噪声对水声信号混沌特征参数的影响。 3．采用RBF神经网络和遗传算法分别建立了两种水声信号的全局预测模型。重点讨论了两种预测模型的学习算法，通过仿真对两种模型的学习速度、所需样本数以及预测效果进行了比较，结果表明RBF神经网络预测模型学习速度快，所需样本点数少，预测效果较好。 4．在建立水声信号预测模型的基础上，提出了一种基于混沌预测的信号检测模型．根据混沌系统的动力学特性，给出了这种检测模型的检验准则。通过对Lorenz信号和实际水声信号进行仿真，验证了检验准则的有效性。比较两种预测模型在仿真中的检测效果，表明预测模型的预测误差越小，所能检测出的信号的信噪比越低。 5．采用支持向量机理论，初步研究了水声信号的分类识别算法。选取两类水声信号，计算它们的关联维数和h₂熵，每类信号各提取32组数据。取两类水声信号各8组数据作为训练样本，训练支持向量机，其它样本用于验证。结果表明，支持向量机的分类算法分类效果较好，比较适合小样本、非线性分类。更多还原

【Abstract】 The detection of underwater acoustic signals has become so difficult with the complexities of the sea ambient as well as the improvement of the ship’s noise reduction technology that it has obvious limited the application of traditional detection methods to underwater acoustic signals. Following the development of chaos theory in recent years, some people have studied the application of chaos theory to the field of underwater acoustic signal processing. For example, some people researched the mechanism of the underwater acoustic signals, prove the signals that have chaotic features, and some of them extracted the characteristic parameters. In this paper, we studied the chaotic feature extraction algorithm, noise reduction algorithm, detection and target recognition algorithms of underwater acoustic signals based on the chaotic theory. The main contributions of the paper are as follows:1. Focused on the research of two methods of choosing the neighborhood size parameter for the de-noise algorithm for underwater acoustic signals, a nonlinear local projective noise reduction algorithm is studied. The methods are based on recurrence plot and noise level estimation respectively. We also compared the noise reduction results of the Logistic, Lorenz and underwater acoustic signals with different noise level by the two parameter selection methods. The results show that the neighborhood size determined by noise level is superior to that determined by recurrence plot.2. The nonlinear characteristic parameters of the underwater acoustic signals were calculated. And after comparing the difference of nonlinear parameters between filtered and un-filtered underwater acoustic signals, the noise influence on the nonlinear characteristic parameters is concluded.3. Two chaotic time series prediction models were proposed. One based on RBF neural network and another on genetic algorithm (GA). After comparing the two models’ learning rate, the number of training set and the forecast effect are concluded. The results show that the RBF neural network model is better than the GA model.4. After building up the two prediction models, a signal detection model based on chaotic forecast theory is explored. An examination criterion is proposed firstly for the detection model. We simulate the detection model by the Lorenz and underwater更多还原