【作者】 刘拥军；

【导师】 葛德彪；

【作者基本信息】 西安电子科技大学 ， 无线电物理， 2010， 博士

【摘要】 合成孔径雷达(Synthetic Aperture Radar, SAR)具有全天候、全天时成像、穿透性强等特点,广泛应用于军事目标侦察、自然灾害监测、资源勘察等领域。其图像信息提取多年以来一直被人们研究,是SAR图像解译(目标识别)的关键步骤,在探测、成像、SAR图像解译中扮演着重要的角色。与SAR数据源的快速增长相比,SAR图像解译技术的研究相对比较滞后。本文的研究重点是舰船、飞机和车辆等目标的特征提取、特征仿真和目标自动识别。根据目标特征提取和目标自动识别的研究过程,在总结前人主要成果的基础上开展了以下工作：1)对SAR图像的特点以及SAR图像中不同类型目标的特征进行了分析,详细讨论了SAR图像中面目标的纹理信息和常见的纹理分析方法。2)分析了SAR图像预处理中的斑点噪声特征,比较了各种消斑方法的优缺点。斑点噪声的存在影响了SAR图像的质量,本文采用小波阈值去噪的方法降低了SAR图像中的斑点噪声,利用Daubechies小波对图像进行2D小波变换,在阈值处理时采用了分层阈值算法和软-硬阈值折衷的方法。实验证明,该方法可以达到较好的去噪效果,并能保留边缘等细节信息,有助于下一步的SAR图像目标特征、纹理特征提取工作。3)研究了Gabor滤波器组在纹理特征提取中的应用。在详细分析了Gabor滤波器的构成和参数选择的基础上,构造了多尺度、多方向的Gabor滤波器组,对要处理的SAR图像进行滤波得到了一系列不同尺度和不同方向上的滤波图像。把提取滤波图像中的局部极值点作为反映目标信息的特征点,用特征点附近的小邻域能量之和表征该点所在区域信息。这是对高分辨率SAR图像中舰船类大型目标利用纹理特征进行目标识别的尝试。实验表明,不同目标的特征提取结果有较大差异,通过合理地设置阈值,可作为区分不同目标的方法,这种特征可应用于后续的SAR图像目标分类和目标识别。4)重点讨论了基于物理光学和几何绕射理论的散射中心理论模型,对散射中心模型的各个参数在图像域进行了估计,详细分析了基于散射中心理论的SAR图像目标特征提取算法,开发了SAR目标自动识别系统软件,采用了先估计目标方位后识别目标类型的目标识别方法以提高目标识别的效率,利用Delaunay三角化技术提高了目标方位的估计精度。实测MSTAR (Moving and Stationary TargetAcquisition and Recognition) SAR图像中目标的识别结果表明了该方法的准确性和有效性。5)实现了从目标建模、散射计算到雷达图像仿真的全过程。首先使用AutoCAD软件对目标建模并进行三角面元离散,采用反方向光学射线追踪技术进行消隐,以物理光学近似理论计算目标的后向散射,最后利用逆合成孔径雷达成像方法得到了目标的雷达仿真图像。球体的计算结果验证了物理光学的正确性,坦克和飞机模型的仿真成像结果与实验结果和CAD模型的对比进一步证实了目标仿真成像系统的有效性。6)利用欧洲非相干散射雷达的常规电离层探测数据,分析了实测数据中接收功率“变”的现象。高度发生在500km以上的功率“突变”一般不是由于地球物理的影响产生的,而是归因于卫星或空间碎片。采用匹配滤波方法实现了对空间碎片的探测和提取,利用相干积累的方法提高了信噪比,共获得了363个碎片,得到了碎片随高度和时间的分布。更多还原

【Abstract】 Synthetic Aperture Radar (SAR) can observe all-weather, all-day and penetrate some objects. Therefore, it is widely applied in many areas, such as military objective reconnaissance, natural disaster monitoring, resource survey and so on. The information abstraction from the SAR images has been studied all the while, which is the key step of the SAR image interpretation and plays the important roles in the detection, imaging and SAR image interpretation. But it is difficult for the interpretation of SAR image to meet to the rapid growth of the data collection capability。This dissertation focuses on the study of the target feature extraction, feature simulation and automatic target recognition (ATR), such as a marine, an aircraft, a vehicle and so on.Following the processes of target feature extraction and SAR ATR, and on the basis of previous studies of many pioneers in this field, some investigations are done in this paper as follows:1) The basic characteristic of SAR image and the features of some typical objects were studied first, especially the texture information in SAR image and some common texture analysis methods.2) The feature of speckle noise in SAR image is analyzed, several speckle removal methods are carried out to reduce the speckle which can corrupt the useful information in the SAR image. The wavelet thresholding method was used to remove speckle noise in the SAR images. In this paper, Daubechies base4 is chosen to make SAR images 2D wavelet transformation. A tradeoff method between soft-thresholding and hard-thresholding was used and the threshold values were calculated at different levels. The experimental result of removing speckles in SAR images was satisfied, and a lot of useful detail information such as edges can be retained.3) A Gabor filter bank in different scales and directions was constructed based on detailed analysis of the structure of the Gabor filter and the relations between the filter parameters. A series of filtered images were obtained by the Gabor bank and the local maximum points were extracted from the filtered images. The feature points were characterized by the sum of the energy values in their small neighboring area around them. This method is an attempt at a marine recognition using high resolution SAR imagery. The differences between the feature points extracted from different objects are obvious, and they can be used as important features for distinguishing one object from the other objects. The Gabor texture feature can be applied to object classification and recognition in some situations.4) The theory of attributed scattering centers based on the Physics Optics(PO) and the Geometric Theory of Diffraction(GTD) is analyzed and discussed. The detailed formula is derived. The parameters of the model are been well estimated on regions of the SAR image. Leading surface estimation (orientation of the target estimation) and Delaunay walk are introduced to improve performance of the SAR ATR system. The recognition results using MSTAR show that the algorithms for estimating the parameters are efficient and the theory of attributed scattering centers is practical.5) The method of radar target modeling, electromagnetic scatter calculation and radar imaging simulation is described. First, it models the object with triangular element by using AutoCAD, reverse ray tracing technique is used to find object shadow, radar scattering character is calculated with the Physics Optics theory, then a high-resolution ISAR image of the target is obtained. The results show that the algorithms for calculating RCS are efficient and the ISAR imaging simulation system is practical.6) Based on the conventional ionospheric data obtained by European incoherent scatter radar, the "mutation" phenomenon of the sounding data power was studied. Generally, the "mutation" power in more than 500km is not produced as geophysical effects, but due to satellites or space debris. Matched filtering method was used to achieve the space debris detection and the debris’s character extraction, the signal to noise ratio is improved by using coherent accumulation method. A total of 363 pieces were captured, and the distribution of debris with height and time has been gotten.更多还原