SAR图像质量评估及其目标识别应用.pdf

中国科学技术大学博士学位论文SAR图像质量评估及其目标识别应用姓名：张倩申请学位级别：博士专业：信号与信息处理指导教师：刘政凯；张荣2010-10-23摘 要 I 摘 要 摘 要 SAR 在军事侦察和民用方面都具有重要作用，应用越来越广泛，因此开展SAR 图像应用的工作具有实际意义和应用前景但是由于 SAR 图像的特殊性导致了 SAR 图像存在一些特殊的质量问题，使 SAR 图像的解译比较困难，最终影响 SAR 图像的应用 因此， SAR 图像质量评估及其应用的研究具有重要的意义 本文首先研究了 SAR 成像的特点，从几何、辐射、统计分布等角度分析了SAR 图像的一些特性然后从图像获取的角度选择了五个客观参数对 SAR 图像进行质量评价，包括均值、方差、等效视数、辐射分辨率和灰度分辨率，高效、直观的对 SAR 图像的质量进行初步评估， 并对真实 SAR 图像的数据库进行了实验验证从图像内容的角度，针对有增益问题的 SAR 图像，提出了一种基于归一化行/列均值的算法进行增益检测，能够快速有效的检测出有增益问题的 SAR图像，并标示出增益变化的位置，有助于对 SAR 系统的改进进行指导。

提出了一种基于相似性度量的高分辨率 SAR 图像无监督分割算法首先以JSEG 算法为基础， 针对 SAR 图像的特点， 以减小斑点噪声对于分割影响为目的，提出了一种新的相似性度量标准SAR 图像经过预处理，得到一个灰度类图，利用该相似性度量标准对此灰度类图进行纹理组合， 建立一个能够反映区域内部和边界的新图在图中，数值低的部分代表区域的中心，数值高的代表区域的边界，利用这个特性，用区域生长算法对新图进行分割最后实验证明该算法从目视解译上更为精确，对于目标相对较弱的 SAR 图像优势尤为明显；而且计算复杂度低，快速、高效，利于实际应用 针对 SAR 图像统计特性，提出了一种基于局部中值拟合 C-V 模型的新的SAR 图像分割算法：LMFCV-SIS该算法核心是利用像素点及以其为中心的邻域内的像素点的局部中值拟合来构造能量函数，极小化该能量函数，得到轮廓的最终演化结果通过一系列对比实验，结果表明，该算法充分利用了 SAR 图像的特征信息，对真实机载 SAR 图像进行分割具有分割边界定位准确、收敛速度摘 要 II 较快等优势 最后建立了一个 SAR 图像的质量评估和目标识别应用验证系统该系统能够对 SAR 图像从不同的角度进行质量评估，包括客观评价、增益检测、重影检测以及模糊度检测等；同时能够进行以识别为主的各种应用，包括道路识别以及河流和桥梁识别等。

利用该系统，对数据库中的 SAR 图像进行了大量的实验，综合质量评估以及一些应用的结果进行分析，即能很好的理解 SAR 图像，并能够对 SAR 系统的研制、改进以及数据的获取提供依据 关键词：关键词：合成孔径雷达 SAR 图像 质量评估 分割 目标识别 Abstract III ABSTRACT ABSTRACT SAR plays an important role in military reconnaissance and civil activity, and SAR images are more and more widely used. Then it is of practical significance and broad prospect to carry out the research on SAR image application. However, due to the special nature of SAR images, there are some characteristic quality issues in SAR images, which make its interpretation more difficult and affect its application ultimately. Therefore, research on SAR image quality assessment and its application is of great significance. In this paper the features of the SAR imaging are studied first, and the characteristics of SAR images are analyzed from the standpoint of the geometry, radiation, statistical distribution and so on. Then from the perspective of image acquisition, five objective parameters are selected to do quality assessment on SAR images, including the mean, variance, equivalent number of looks, radiometric resolution and grayscale resolution. It is verified by experiments on the real SAR image database that the initial quality assessment of SAR images using these five parameters is efficient and intuitive. From the perspective of image content, for SAR images with the gain problem, a gain-detecting algorithm is proposed based on the normalized row/column mean, which can detect effectively ones with the gain problem from SAR images and mark the location of the gain changes on these images. This can help to give suggestions on the improvements in SAR systems. An unsupervised segmentation algorithm for high-resolution SAR images is proposed based on homogeneity criterion. According to the characteristics of SAR images, a novel homogeneity criterion based on JSEG (J-Segmentation) was proposed to reduce the impact of speckle noise on the segmentation. After preprocessing, a gray class-map is created. Using the homogeneity criterion, the gray class-map is mapped to a new data which can indicate where are region centers and where are region boundaries. In the new map, low and high values correspond to possible region centers and region boundaries. According to this feature, the new map is segmented with region growing. The performance of the approach is verified that the algorithm is more accurate in line with a visual interpretation especially for those SAR images with relatively weak targets, and this algorithm is fast, efficient and beneficial for practical application and it has lower computational complexity. Abstract IV A novel SAR image segmentation algorithm based on Local Median Fitting C-V model (LMFCV-SIS) is proposed according to the characteristics of SAR images. The main idea of the algorithm is that the LMF of the pixel and its neighbors is used to form an energy and the final evolution of the curve is given by the minimization of the energy. The performance of the approach is verified by plenty of real airborne SAR images and the experimental results on the real data show its efficiency and accuracy. Finally, an authentication system is established for quality assessment and target recognition of SAR images. Using the authentication system the quality of SAR images can be assessed from different angles, including objective assessment, gain detection, ghost detection, blur extent evaluation and so on. The system can also be used for doing a variety of applications, including road recognition, river recognition, bridge recognition and so on. In this system, a large number of experiments are carried out on SAR images in the database. Using the comprehensive analysis of the results of quality assessment and those applications, SAR images can be well understood, and the basis can be given for the development and improvement of SAR systems and SAR data acquisition. K。