【作者】 段崇雯；

【导师】 胡卫东；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2013， 博士

【摘要】 海面舰船类别属性的判断，对国家的海防安全和航运交通管理具有重要的意义。合成孔径雷达（Synthetic Aperture Radar，SAR）由于具有全天时、全天候的工作能力和精细的分辨能力，在舰船目标的分类和识别中发挥了巨大作用。利用SAR成像提取的舰船目标特征参数，是对其进行分类识别的重要依据。本文针对星载/机载SAR对海侦察监视的应用背景，从四个方面对高分辨率成像条件下舰船目标特征参数估计技术开展了研究。考虑到研究中能够获得的实测舰船SAR图像在数量和质量上的不足，采取主要以电磁散射建模仿真数据开展算法研究，并辅以实测图像验证的研究途径。因此，首先对现有的电磁散射建模方法和典型的建模软件进行了概括总结，并对本文所采用的舰船-粗糙面复合模型以及所采用的混合建模技术进行了说明。为了验证该研究途径的可行性，利用几个模型算例，将仿真的雷达散射截面积和SAR图像与经典建模软件的仿真结果以及实测数据成像结果进行了比较。其次，针对海面舰船目标SAR图像中存在噪声和海浪杂波干扰的问题，研究了以Capon谱估计为主要手段的SAR图像预处理技术。将Capon方法与其他常用谱估计技术进行了比较，并且利用Monte Carlo仿真分析了该方法在提高散射中心分辨率和图像目标-杂波比方面的性能。同时，提出一种迭代式的实现方法，缓解了传统二维Capon谱估计内存消耗过大的问题。进一步地，针对较低目标-杂波比的图像，为提高姿态角估计的稳健性，提出了基于二值图像Radon变换角度熵的姿态角估计技术。第三，为了研究不同海况条件下由海面反射引起的电磁波多径传播效应对舰船图像及其特征参数估计的影响，在海面满足基尔霍夫近似条件并具有特定波高-斜率联合概率密度函数的假设下，建立了粗糙面上方散射中心雷达回波中多径延迟量的概率模型。在实验部分，分析了该概率模型与散射中心高度，雷达视角，以及海面特征参数之间的关系。作为应用，分别将其用于二维和三维场景下海面舰船目标雷达成像的特征参数估计，并结合海况条件给予不同的处理。最后，针对从SAR图像中提取的舰船区域长度、宽度和姿态角等二维几何参数存在不确定性，而基于散射中心的三维重构在舰船像存在散焦时又难以开展的问题，给出了一种基于轮廓的舰船目标三维几何特征估计方法。该方法以多幅已知俯仰角的舰船目标SAR图像为基础，利用特征椭圆概括出舰船像的二维几何特征，然后通过重构这些椭圆的三维目标椭球，估计出舰船的三维尺寸和成像雷达的方位角。在实验部分，利用Monte Carlo仿真对影响算法性能的主要因素进行了分析，然后应用多个不同类别舰船模型的仿真SAR图像开展特征参数估计和目标分类实验，对算法的有效性进行了验证。更多还原

【Abstract】 Ship target identifiation is of great importance for the national coastline defenseand the shipping traffic management. With the fine resolution in all-day, all-weatherconditions, Synthetic Aperture Radar (SAR) plays an important role in the classificationand recognition of sea ships. To achieve this goal, a variety of ship features areextracted from SAR images. This dissertation, against the background of sea surfacereconnaissance and surveillance by using high resolution spaceborne/airborne SAR,focuses on the ship feature parameter estimation in the following four aspects.As regarding to the inadequacy in both quantity and quality of measured images,this dissertation uses simulated ones generated by the methods of electromagnetic (EM)modeling for most experiments, which are followed by some validations with measuredimages. For the purpose of self-containedness, some typical techniques of EM modelingand popular softwares are introduced. The composite models of the ships and the roughsurfaces used in this dissertation are then illustrated, as well as the hybrid method of EMmodeling. To testify the validity of the applied approach, the simulated radar crosssection and SAR images are provided and compared with those generated by the classicsoftware and the real measured data.Secondly, in order to suppress the disturbance of the noise and sea clutters in SARimages, the image pre-processing based on the Capon spectral estimation method isstudied. The Capon’s method is compared with other spectral estimation techniques.Subsenquently, its performances in improving the resolution of scattering centers andthe Target-to-Clutter Ratio (TCR) of SAR images are examined through the MonteCarlo simulations. Moreover, to solve the severe problem of memory comsuption, aniterative way to realize the two-dimensional (2D) Capon is put forward. Meanwhile, forthe SAR images with low TCR, a robust pose estimation algorithm is presented basedon the angle entropy of the Radon transform of the bi-valued images.Thirdly, to study the effect of multipath propagation on the ship image and featureparameter estimation under different sea conditions, a probability model is deduced forthe radar echo multipath delays of the scattering centers on a rough surface. In thismodel, the sea is considered as a rough surface satisfying the Kirchhoff approximationand described by a certain elevation-slope probability density function. Experimentally,we analyzed the relationship between the model and the factors such as the scatteringcenter height, the radar elevation angle, and the sea characteristic parameters. Themodel is then used in the feature parameter estimation in both2D and three-dimensional(3D) problems, with different modifications for different sea conditions. In the end, to eliminate the uncertainty in the2D geometrical features, such as thelength, width, and pose angle, an outline-based method is proposed to estimate the3Dfeatures for ship images with unfocused scattering centers. The method utilizes morethan one image with known radar elevation angles and extracts the feature ellipses ofthe ship outlines. By reconstructing the objective ellipsoid of these ellipses, one canfinally get the3D geometrical features of the ship, including the size in each dimensionand the radar azimuth while imaging. The performance of the algorithm is analyzedwith Monte Carlo simulations. Moreover, feature estimation and target classificationexperiments are carried out with simulated images of several kinds of ships, whichtestify the effectiveness of the algorithm.更多还原