【作者】 王福友；

【导师】 袁赣南；

【作者基本信息】 哈尔滨工程大学 ， 导航、制导与控制， 2009， 博士

【摘要】 海杂波(海表面回波),即来自雷达发射脉冲照射局部海面的后向散射回波。为了深入揭示海杂波的内在物理特性和规律,在实测IPIX雷达海杂波数据基础上对海杂波进行混沌分形特性分析、建模,同时为了解决传统舰载和岸基雷达难以检测出海杂波背景下小目标的现状,在本文研究出的海杂波的内在物理特性和统计规律基础上对海杂波背景下小目标进行检测。论文主要研究工作如下：1.在实测海杂波数据基础上,运用关联维、Lyapunov指数、Kolmogorov熵三个指标判断出海杂波的时间序列的混沌特性。针对海杂波的电磁散射是一种时空现象,提出研究海杂波的时空混沌特性更能揭示海杂波的本质,基于实测数据对海杂波的时空混沌进行定性分析,运用关联长度和借助耦合映像格子(CML, Coupled Map Lattice)模型计算最大Lyapunov指数判断出海杂波具有时空混沌特性。2.对海杂波分形特性进行较深入地研究。基于小波分析对海杂波的Hurst指数进行了计算,得出距离单元含有小目标和只含有海杂波的Hurst指数具有明差异的结论。考虑到海杂波的多奇异性和非线性非平稳的特点,提出基于小波模极大值(WTMM, Wavelet Transform Modulus Maxima)和多分形消趋势波动分析(MFDFA, Multifractal Detrended Fluctuation Analysis)两种方法对海杂波进行多重分形特性进行分析,分别计算了多重分形指标：尺度指数τ(q),广义Hurst指数h(q),奇异谱f(α)三个指标,得出单一距离单元的海杂波时间序列具有多重分形特性。为了进一步分析一组数据多个距离单元的海杂波分形特性,提出一种时间-距离单元-幅度图法对其进行分析,得出一组数据的多个距离单元的海杂波数据也具有分形特性,尤其是数据中含有小目标更佳表现多重分形特性,这进一步说明海杂波和小目标的分形特性不同。3.针对海杂波是一种时空现象和海杂波非线性非平稳的特点,提出一种基于最小二乘支持向量机-耦合映像格子(LSSVM-CML, Least Squares Support Vector Machines-Coupled Map Lattice)的海杂波时空预测方法,实验表明该方法的预测效果优于基于加权一阶局域法、归一化RBF神经网络、Volterra预测器、LSSVM算法的海杂波时间序列非线性预测效果。4.为了解决传统雷达恒虚警率(CFAR, Constant False Alarm Rate)方法不能有效地检测出海杂波背景下小目标的现状,在海杂波的时空混沌特性、多重分形特性、时频分析和统计规律基础上提出4种新的小目标检测方法：LSSVM-CML方法、广义分形维偏差法、时间-Doppler法、局部幅值统计法,实验结果表明这4种方法是有效的,可在对小目标及海洋环境没有先验知识条件下较准确地检测出小目标。5.针对实测数据存在的雷达测量噪声和来自于粗糙海面的动态噪声,这些噪声会影响海杂波内在特性分析,因此论文采用均值、中值、小波阂值法、经验模式分解(EMD, Empirical Mode Decomposition)算法分别对海杂波信号进行去噪,重点研究了基于小波阈值法和EMD算法的海杂波信号进行去噪。实验结果表明db2小波双曲线阈值函数HeurSure阈值模式和EMD算法去噪效果较好,由于EMD算法去噪是全数据驱动的,因此去噪效果最优。更多还原

【Abstract】 "Sea clutter", refers to radar backscatter from local sea surface. In order to disclosure the intrinsic physical characteristics and laws of sea clutter, chaotic and fractal analysis and modeling is performed based on the real-life IPIX sea clutter datasets. Meanwhile, aiming at the present issue that small target detection within sea clutter is very difficult for traditional and coast-sited radar, investigation of small target detection is achieved based on the chaotic and fractal characteristics and laws of sea clutter obtained from this dissertation. This thesis concentrates on as follows:1. Based on the real sea clutter data, the sea clutter is regarded as chaotic by calculating Correlation dimension, Lyapunov exponent and Kolmogorov entropy. Considering radar electromagnetic wave scattering from a rough sea surface is basically a spatiotemporal phenomenon, this dissertation presents that the study of spatiotemporal chaos (STC) can further disclosure the true nature of sea clutter, STC qualitative, analysis is achieved by real sea clutter data, STC quantitative analysis is achieved by calculating Correlation length and the largest Lyapunov exponent via Coupled Map Lattice (CML) model.2. Fractal characteristic of sea clutter is deeply investigated. Wavelet analysis is applied to calculate Hurst exponents of sea clutter, it is found that the Hurst exponent of the range bin hosting a small target is different from hosting only sea clutter. Considering dense singularities and nonlinearity and nonstationary of sea clutter, Wavelet Transform Modulus Maxima (WTMM) and Multifractal Detrended Fluctuation Analysis (MFDFA) are derived for multifractal analysis of sea clutter, scale exponent r(q), generalized Hurst exponent h(q), singularity spectrum f(a) are calculated, the results indicate that sea clutter time series of a certain range bin is multifractal. For further study of the fractality of a sea clutter dataset with many range bins, a new method of Time-Range bin-Amplitude Plot (TRAP) is presented, experiment results imply that a sea clutter dataset with several range bins has fractality characteristic, especially when the dataset contain a small target, it performs multifractal characteristic, which further implies that sea clutter and small target has different fractality.3. As to sea clutter is a spatiotemporal phenomenon and nonlinear and nonstationary, a new algorithm of Least Squares Support Vector Machines-Coupled Map Lattice (LSSVM-CML) is presented for spatiotemporal prediction of sea clutter. Experiments results indicate that the prediction precision of LSSVM-CML algorithm is superior to Weighted One-order Local-region algorithm, Normalized RBF Neural Network, Volterra Predictor, LSSVM algorithm for prediction of sea clutter time series.4. In order to solve the problem that the traditional radar Constant False Alarm Rate (CFAR) method can not detect small target in sea clutter, based on the STC characteristic, Multifractal characteristic, time-frequency analysis, statistical regularities obtained in this dissertation, four new methods including LSSVM-CML algorithm, Generalized Fractal Dimension Difference (GFDD) algorithm, Time-Doppler analysis (TDA) and Local Amplitude Statistics (LAS) method are derived for small target detection within sea clutter. Experiments results indicate that these methods are valid, the small target can be detected precisely with no prior knowledge of the small target and ocean environment conditions.5. Since the real-life sea clutter data contains radar measurement noise and dynamic noise from rough surface which may influence the analysis of the intrinsic characteristics of sea clutter, so Averaging, Median, Wavelet and Empirical Mode Decomposition (EMD) algorithm are used for denoising. Wavelet thresholding method and EMD algorithm for denoising is the major investigation in this dissertation. Experiments indicates that db2 wavelet with Hyperbolic Thresholding function and HeurSure threshold and EMD algorithm for denoising of sea clutter are excellent, since EMD-based signal denoising is full data driven, so EMD algorithm denoising is optimist a little better than Wavelet thresholding denoising method.更多还原