【作者】 夏昕；

【导师】 罗代升；

【作者基本信息】 四川大学 ， 电路与系统， 2006， 硕士

【摘要】 合成孔径雷达(Synthetic Aperture Radar，SAR)是一种工作在微波波段的成像系统，具有全天侯、全天时、多视角、多分辨率数据获取的能力。因此，它在国防、军事、环境、灾害等的侦察和监视中有着不可估量的应用价值。 合成孔径雷达成像应用中最重要、最难、最热门的研究课题之一是自动目标识别。迄今为止，还没有一种方法可以解决所有的自动目标识别问题。因此雷达目标识别的研究具有重要的理论意义和学术价值。 本研究的目的旨在研究典型目标的特征提取与识别方法，提高目标识别率。 本文提出了基于堆积几何特征的SAR图像典型目标识别方法。由于油罐是一种典型的高散射系数金属硬目标，在低分辨率情况下，合成孔径雷达图像中油罐个体的散射集中体现为高亮度的点，待识别的各点状目标在场景中堆积为呈一定分布面积的目标区域，目标中心呈直线规则分布，相干斑噪声零散分布，因此利用点状目标与噪声不同的灰度分布与空间几何分布特点，采用该方法解决了这类低分辨率下密集型点状目标的识别问题。 本文提出了基于边缘几何特征的SAR图像典型目标识别方法。由于在高分辨率情况下，合成孔径雷达图像中油罐个体的散射分别体现为由若干高亮度点围成的椭圆，待识别油罐目标的模糊边缘呈不连续的椭圆状，因此从二值化后的SAR图像中识别出拟目标的不连续边缘，判断其是否满足目标边缘特性来识别出最终结果，采用该方法解决了这类高分辨率下椭圆状目标的识别问题。 本文运用了基于Gauss核函数支持向量机的SAR图像典型目标识别方法。由于上述基于边缘几何特征的识别方法并未考虑目标的统计差别，故本文又从目标统计特征的角度，根据训练样本集求出支持向量，构造出最优分类面方程，实验证明最优分类面相对于训练样本为最优的，不同的训练样本得到的最优分更多还原

【Abstract】 SAR is a kind of coherent radar imaging system working in microwave bandwidth. It has the capability of imaging in all weather, all day, multi-view angles and multi-resolutions. Thus, it has undescribable values of applications in spy and monitoring of denfence, military, environments, and disasters.Automatic target recognition is one of the most important, difficault, and hottest research subjects in SAR imaging applications. So far, there is not a single method which can solve all the target recognition problems. The research on recongnition, therefore, is significant and baluable in theories and technologies.The aim of this research project is at development of methods for feature extraction and recognition of typical targets to promote the recognition rates.In this dissertation, a recognition algorithm based on the geometry feature of gather is proposed to recognize the typical target in SAR image. Oilcan is a kind of typical rigid target with big backscattering coefficient, which takes on point figure if the resolution of SAR image is not very high; these point targets to be recognized gather together, and form a target region with some distribution area; the center of the region appears as a beeline, and the speckle noise appears scattered, therefore, considering the different grey distributions between point target and speckle noise region and the space geometry distribution, these point targets can be recognized by this algorithm.A recognition algorithm based on the geometry feature of the edge is proposed更多还原