合成孔径雷达因其全天候和高分辨率成像的特点，近年来得以快速发展。与此同时，随着近代的计算机技术和先进的数字信号处理技术的发展，合成孔径雷达在许多领域得到广泛的应用，如大地遥感和测绘、大地资源探测、灾情预报、军事指挥以及国民经济的其它方面。由于SAR相干成像方式使得SAR图像中存在显著的相干斑噪声，降低了图像质量，影响了后续目标检测、分类和识别等应用，因此SAR图像的相干斑抑制一直是合成孔径雷达技术领域重要课题之一。 支持向量机(Support Vector Machines，SVM)是Vapnik等人基于统计学理论基础上提出的新一代学习系统的一类新型机器学习方法。由于其出色的学习性能，该技术已成为机器学习领域的研究热点，并在实际当中得到了一定广泛的应用，如人脸检测、手写体数字识别、文本分类等。SVM作为一种新兴技术，在很多领域的研究有待探索和完善。 本文针对雷达数据的特点，首先将图像数据通过支持向量回归滤波网处理，达到最大限度保留边缘图像信息同时滤除噪声，然后根据分类理论与技术，研究某一类特定目标的分类方法。 主要研究内容有： (1)本文分析去除相干斑的一些基本算法，深入研究了利用支持向量回归技术构造滤波网问题，并将滤波网模型应用在SAR图像增强处理中。通过实验证明支持向量回归滤波网络在SAR图像增强方面的实用性，为下一步的目标分类工作提供了基础。 (2)通过分析核函数与SVM分类器性能之间的关系，选择组合核函数构造SVM分类器，并将其运用在SAR图像分类中。大量的对比实验验证，利用组合核函数构造的SVM分类器，不但从性能上改善分类器的效果，而且明显的提高了SAR图像的分类精度。

In recent years synthetic aperture radar gets to the rapid development, because of its high-resolution and all-weather characteristics. At the same time, modern computer technology and advanced digital signal processing technology had developed and SAR made wide applications in many areas, such as earth remote sensing, mapping, and the exploration of resources, and disaster forecasting, military command, as well as other aspects of the national economy.The coherent imaging method allows speckle noise in SAR images ,the speckle noise reduce the image quality and have an impact on the follow classification and identification applications, so the Speckle Reduction of the SAR image has been an important subject of synthetic aperture radar technical fields. Vapnik and others made a new generation system of machine learning methods on the basis of statistical theory. for excellent performance ,SVM has become the hot in the machine learning research , and in many areas have been successful applications,for example ,face detection, handwritten numeral recognition, text classification. As a new technology, SVM has further study in pattern recognizing. With the characteristics of radar data, first, the image must to through regression network processing in order to keep the edge image maximizing information and filtering noise , then according to the classified theory and technology work over classification method of particular type. Major study: (1)This paper analyses some of the basic speckle removal algorithm, In-depth study the support vector regression filtering technology network structure issues, and filtering network model used in SAR image enhancement processing. Through experiments proved that support vector regression filtering network’s Practicability in the SAR image enhancement, for the next work provides a basis theory for classification. (2)To investigate the connection with kernel and SVM classification,choiced compounding kernel as SVM classification kernel , and used in SAR image classification . A large number of comparative experiments show that the compounding kernel SVM classification, not only improved performance of the classification, but also enhanced the accuracy of the classification in SAR images.