【作者】 李丽亚；

【导师】 吴顺君；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2009， 博士

【摘要】 逆合成孔径雷达(ISAR)图像是目标在距离-多普勒平面上的投影,包含了目标的二维结构信息,基于ISAR图像的目标识别是宽带雷达自动目标识别领域的重要组成部分。极化是描述目标散射特性不可或缺的一部分,极化和宽带高分辨技术相结合是宽带雷达自动目标识别极具潜力的发展方向。因此,本论文围绕ISAR雷达目标识别和极化雷达目标识别这两个问题对干涉ISAR(InISAR)三维成像、干涉ISAR目标识别、多极化雷达高分辨一维距离像(HRRP)识别和基于极化散射矩阵(PSM)的极化雷达目标识别进行了分析和研究。本论文的主要内容概括如下:1.分析了干涉ISAR三维成像的原理,利用了目标体上强散射点处信噪比较高的特性,提出了一种新的基于特显点的干涉ISAR成像方法,该方法能有效地解决相位解缠绕问题和在低信噪比传统干涉ISAR成像方法无法正确成像的问题;并且对斜视情况下的干涉ISAR成像问题进行了研究。2.对干涉ISAR三维成像中的一些问题进行了分析和研究。第一个问题是干涉ISAR三维成像中的天线布阵方式和基线的设置,包括两种布阵方式、横向分辨率与基线的关系、基线去相关和最优基线;第二个问题是角闪烁,分析了角闪烁产生的机理、特征和解决办法;最后一个问题是超分辨算法在干涉ISAR成像中的应用,给出了两种经典的超分辨干涉ISAR成像方法。3.提出基于干涉ISAR像的雷达目标识别方法。给出了干涉ISAR图像的预处理方法,接着使用极化映射提取具有旋转不变性和尺度不变性的特征,最后设计三种分类器对目标进行识别。对影响干涉ISAR成像和识别的四个因素:俯仰角度、目标转速、天线孔径(也称基线)和目标与雷达的距离进行了分析,实验结果验证了理论上的分析。4.多极化HRRP包含了比单极化HRRP更多的目标结构特征信息,因此采用多分类器融合多极化信息可以提高雷达目标识别性能。提出了基于度量层的特征加权融合算法和基于决策层的加权投票融合算法,它们使用不同的准则对多极化HRRP进行融合分类,取得了较好的识别性能。5.将核方法引入多极化HRRP识别中,提出了两种基于多极化HRRP的核函数,将这两种核函数应用到核主分量分析(KPCA)中提取特征,进而使用分类器对目标进行识别。该方法可以在不丢失极化信息的情况下,将多极化HRRP作为一个整体进行识别,降低了识别算法的复杂度。6.针对基于PSM的目标识别中极化特征提取困难的问题,提出了基于核函数的识别方法。定义了两类基于PSM的核函数,将其应用到KPCA中提取特征,对目标进行识别,并取得了良好的识别效果。将提出的两类基于PSM的核函数应用到一种依赖数据的核函数优化中,得到优化的对数据自适应的核函数,提取KPCA特征作为分类器的输入,进而对目标进行分类识别。更多还原

【Abstract】 Inverse synthetic aperture radar (ISAR) images represent the projection of the target onto the range-Doppler plane, which contain the informative 2-D target structure signatures. The target recognition using radar target ISAR images is an important field of the wideband radar automatic target recognition (RATR). Polarization is an indispensable component of the description of the target electromagnetic characteristic. Wideband polarization RATR has received more and more attentions. This dissertation provides our researches for ISAR and polarization target recognition. Interferometric ISAR (InISAR), InISAR target recognition, multi-polarized high resolution range profile (HRRP) target recognition and polarization scatter matrix (PSM) target recognition have been developed in this dissertation. Details are described as follows.1. The theory of InISAR is introduced briefly. Using the characteristic of the higher SNR at stronger pixels, a novel method of InISAR imaging based on the dominant scatterers is proposed. This method can deal with the low signal-to-noise ratio and phase wrapping in InISAR imaging. InISAR imaging in squint model is also researched.2. The issues of InISAR 3-D imaging are discussed. The first issue is about the antenna array and the baseline. Two kinds of the antenna array, the relationship of the azimuth resolution to baseline, the baseline decorrelation and the optimal baseline are presented. The second issue is angle glint. The theory and the suppression are analyzed. The applications of super-resolution in InISAR imaging is the last issue. Capon and Relax method are discussed and compared with FFT method.3. A method of the InISAR target recognition is proposed. The InISAR preprocessing is presented. The features are extracted from the polar image that is obtained from InISAR image by the polar mapping. The extracted features have invariance with respect to rotation and scale. The effects of the four important parameters (elevation, speed, baseline and range) on imaging and recognition are discussed, the results of four experiments prove the theory analysis.4. The multiple-polarized HRRP includes much more target information than single-polarized radar HRRP dose, so using multiple classifiers to combine multiple-polarized information can enhance the radar target recognition performance. Two methods of combining multiple classifiers are proposed, which are the weighted average algorithm and the weighted voting algorithm. Employing the two different combining rules can fuse and classify the multi-polarization radar HRRP. The good recognition performances are achieved.5. Aiming at the great quantity of multi-polarized HRRP, the complexity of the data distribution and the recognition algorithm, the methods based on kernel methods are proposed. Two kernel functions based on the multi-polarization HRRP are defined, then two kernel functions are employed to the kernel principal component analysis (KPCA) respectively. The multi-polarized radar HRRP can be recognized as a whole one in the proposed methods, so the complexity of the recognition algorithm is reduced.6. Aiming at the difficulty of the feature extraction from the polarization scatter matrix (PSM), the kernel methods based on PSM are proposed. Two kinds of kernel functions based on PSM are defined, and they are employed to KPCA respectively. The proposed methods achieve good recognition performance. The proposed two kinds of kernel functions based on PSM are employed to the kernel optimization. Using a data-dependent kernel, an optimized kernel is obtained by maximizing the kernel Fisher criterion. The KPCA features are used as the input of the classifier to classify the targets.更多还原