【作者】 王义敏；

【导师】 安锦文；

【作者基本信息】 西北工业大学 ， 导航、制导与控制， 2006， 博士

【摘要】 作为主动寻的探测设备，合成孔径雷达(SAR)可以全天候、全天时提供高清晰度图像，从而在军事侦察和民用上得到了广泛的应用。但是由于其相干成像的特殊性，因而在SAR图像中包含有大量的相干斑噪声，使得对SAR图像的处理、分析和解译非常困难。而且面对广阔而复杂多变的地形环境，SAR图像背景环境极为复杂，并且SAR图像中含有大量的冗余数据，因此如何综合运用各种知识，对算法的实用性、快速性、有效性和检测精度等性能进行折中是目标检测算法的难点和关键点。本文主要对高分辨率机载SAR图像的目标检测和识别进行研究。主要研究以下几个方面内容： 1、详细分析和研究SAR图像相干斑杂波的统计分布模型和各种统计分布模型的适用性。对几种广泛使用的典型非高斯数据分布模型(Gamma分布、Weibull分布、K分布、有限混合高斯和混合Gamma分布)的数学特点进行了详细讨论和分析，并给出了不同分布模型下参数向量的估计公式。在此基础上，在MSE准则下，对T72坦克的几种不同成像条件下的高分辨率SAR图像数据的统计分布模型(Gamma分布、Weibull分布及有限混合高斯分布)以及对基于Ku波段的机载序列SAR图像数据的统计分布模型(有限混和高斯分布)进行分析和估计。 2、在SAR图像数据统计分布模型研究的基础上，对SAR图像的目标信息自动提取方法进行研究。主要提出了基于局部窗口的恒虚警(CFAR)方法、MAP方法和基于上下文关系的方法。在对单／多T72坦克SAR图像的目标检测方面，这些方法从不同的角度表现出了对SAR图像信息自动提取的能力。并且运用MAP方法对在不同方位角下对太阳能塔所成的机载SAR条带图像序列进行目标检测和分割，仿真结果表明MAP方法对序列SAR图像的目标检测是有效和稳健的。相对于恒虚警方法和贝叶斯方法，基于上下文的SAR图像信息提取方法具有较大的时间消耗外，三种方法均具有较强的鲁棒性和SAR图像的信息提取性能。在SAR图像的目标信息自动检测方面具有明显的优势和应用潜力。 3、提出了一种新的SAR图像的目标区域特征提取和目标识别方法。在基于Hu矩不变特征的基础上，引入了描述目标区域灰度信息的狄度均值和方差系数，将这九个目标区域特征量作为对目标进行识别和分类的依据。对不同分更多还原

【Abstract】 As an active detector, Synthetic Aperture Radar (SAR) which provides high-resolution images in all-weather, all day and all night is widely used in military reconnaissance and civil field. But as a result of the charasteristics of coherent imaging, the speckle is inevitably appeared in the SAR images and makes the processing, analysis and translation of SAR images extremely difficult. And, due to the extensive, complex and dynamic environments, the SAR image backscatter is confused and its data are so redundant. Therefore, how to synthesize various information and get a tradeoff of performances between the practicability, robustness, quickness, effectivity and accuracy of autonomous target detection have become more difficulty and critical. In this thesis, techniques for ground target detection and recognition in high resolution airborne-SAR images are extensively investigated. The main contributions of this dissertation are as follows:1、 The statistical distribution model and their application of the coherent clutter in SAR images are detailedly analysized and investigated. The characteristics of several kinds of widespreadly typical non-Gaussian distribution models, such as the Gamma distribution, the Weibull distribution, the K-distribution, the finite Gaussian mixture distribution and the finite Gamma mixture distribution, are extensively discussed and studied. Then the formulae of estimating parameters’ vector of these models are deduced. Last, using the MSE criterion, we analysis and estimate the statistical models(such as the Gamma distribution, the Weibull distribution and the finite Gaussian mixture distribution) of some different imaging high-resolution T72 SAR image data, and apply finite Gaussian mixture distribution to analyse statistical models of airborne synthetic aperture radar movie of Sandia National Laboratories solar power tower, movie made from multiple aspect images.2、 Based on the statistical models of SAR image data, techniques for extracting target information from SAR images are probed. We proposed the local window-based Constant False Alarm Rate(CFAR), Maximum a Posteriori (MAP) and context-based methods. All these methods show the capability of extracting target更多还原