水下运动目标被动声纳信号建模研究

发布时间：2018-04-20 11:26

本文选题：被动声纳信号 + RBF神经网络　；参考：《昆明理工大学》2014年硕士论文

【摘要】：在水声信号处理领域,水下运动目标被动声纳信号的分析和处理一直是该领域的研究热点,是水声对抗及鱼雷预警技术中的重要环节和关键技术。然而,实测的水下运动目标被动声纳信号微弱,因海水环境及声传播条件不确定而具有时变特性,并夹杂有大量的背景干扰噪声难以提取有效成分。另外,实测数据需要专业设备和人员参与,其耗时长、开销大和数据保密等因素都对被动声纳信号的分析和应用(比如用于海上部队训练、目标指挥系统的方案论证、模拟训练以及对声纳系统在实验室阶段的测试等研究)形成了制约。因此,对水下运动目标被动声纳信号的模拟或仿真有着重要的现实价值和军事意义。本文详细分析了水下运动目标辐射噪声信号(以下简称为原始信号)特性,包括原始信号的数学和统计特性,产生机理等。在此基础上对原始信号中存在的线谱和连续谱分量进行了分析和仿真。针对原始信号在复杂海洋环境传播中受到的多种因素影响模拟了声纳端的被动声纳接收信号(以下简称为目标信号)。由于神经网络具有自组织、自适应和多线程并行处理的特性,以及对于非线性系统良好的逼近能力,本文探索性的使用神经网络技术对原始信号经海水非线性信道传播的过程进行了模拟。并对RBF神经网络学习算法进行了创新性的研究,在深入探讨K-均值聚类和传统FCM算法的基础上,提出了具有更好泛化能力和自适应性的改进型FCM算法,应用此改进算法对信号数据进行聚类分析,确定神经网络结构和基函数参数；通过对非线性函数的逼近和预测仿真,验证了基于本文算法设计的RBF神经网络对于非线性系统拟合的先进性；随后使用本文方法对原始信号传播至被动声纳端的非线性过程进行建模,模拟了目标信号和时域特征值,并运用相关系数和误差指数作为评价指标对模型输出数据和目标信号进行了相似性对比,结果证明所提出的改进算法较经典算法优越,达到了对水下运动目标被动声纳信号建模研究的目的。
[Abstract]:In the field of underwater acoustic signal processing , the analysis and treatment of passive sonar signal of underwater moving target has been a hot spot in this field . It is an important link and key technique in underwater acoustic countermeasure and torpedo early warning technology . However , the measured data requires professional equipment and personnel to take part in the analysis and application of passive sonar signal ( for example , it is used for the training of naval forces , the program demonstration of the target command system , the simulation training and the testing of sonar system in the laboratory stage , etc . ) . Therefore , it has important realistic value and military significance to the simulation or simulation of passive sonar signal of underwater moving target .

In this paper , the characteristics of underwater moving target radiated noise signal ( hereinafter referred to as the original signal ) are analyzed in detail , including the mathematical and statistical characteristics of the original signal , the generating mechanism , etc . On the basis of this , the line spectrum and the continuous spectral components present in the original signal are analyzed and simulated . The passive sonar receiving signal ( hereinafter referred to as the target signal ) of the sonar end is influenced by various factors which are influenced by the original signal in the propagation of complex ocean environment .

As the neural network has the characteristics of self - organizing , self - adaptive and multi - thread parallel processing , and good approximation ability for nonlinear systems , this paper simulates the process of nonlinear channel propagation of the original signal by using the neural network technique , and proposes an improved FCM algorithm with better generalization ability and adaptability based on the deep discussion of the K - means clustering and the traditional FCM algorithm .
Based on the approximation and prediction simulation of nonlinear function , the RBF neural network designed based on this algorithm is validated for nonlinear system fitting .
Then the nonlinear process of propagation of the original signal to the passive sonar end is modeled by using the method , the target signal and the time domain characteristic value are simulated , and the correlation coefficient and the error index are used as the evaluation indexes to carry out similarity comparison on the model output data and the target signal , and the result proves that the proposed improved algorithm is superior to the classical algorithm and achieves the purpose of modeling the passive sonar signal of the underwater moving target .

【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【参考文献】