SAR图像的变化检测方法研究

发布时间：2018-06-30 01:46

本文选题：SAR图像 + 变化检测　；参考：《东北大学》2014年硕士论文

【摘要】：变化检测是从同一地区、不同时相的图像中分析和确定地物变化的特征、过程的技术,是遥感图像研究领域的重要分支。SAR (synthetic aperture radar)技术同可见光成像相比,不受天气和光照条件影响,具有多极化、全天候、全天时和一定穿透性的特点,十分适合变化检测研究。但由于SAR图像中包含相干斑噪声,又使SAR图像区别于其他遥感图像,形成了一套独特的变化检测方法。本文以提高SAR图像变化检测方法的性能为目标,围绕SAR图像的变化检测技术开展研究,主要工作可以概括为以下几个部分：(1)介绍SAR相关原理及变化检测理论,概括SAR图像变化检测的一般流程及基本方法,分析相干斑噪声产生的机理、模型和统计特性,针对SAR图像变化检测的结果给出精度评估的标准和依据。(2)详细介绍Lee、Kuan、Frost和Gamma Map滤波算法原理,分析不同滤波算法的优点、缺点以及适用条件,并通过实验比较,选取Frsot滤波作为本文变化检测研究的滤波算法。(3)比较和分析差值法、比值法和对数比值法构造差异图像的特性,并在此基础上,本文提出用增强的对数比值法来构造差异图像,以提高图像中发生变化和未发生变化区域的可分性。(4)基于图像分割的自动阈值选取算法,对增强的对数比值法构造的差异图像应用循环迭代法、最大类间方差法和最佳直方图熵法；基于贝叶斯决策的自动阈值选取算法,对增强的对数比值法构造的差异图像应用KI最佳阈值选取算法和EM迭代阈值选取算法。分析了传统算法的优缺点及限定条件,针对以上算法的不足之处,首次将基于最小Tsallis交叉熵的图像分割算法应用到SAR图像的变化检测,平均正确检测率达96.1916%,平均Kappa系数达0.58,提高了变化检测性能。(5)提出改进的基于最小Tsallis交叉熵阈值选取算法,充分利用图像中每个像元与周围像元的空间关系,对基于最小Tsallis交叉熵选取的单一阈值进行补偿,用获得的动态阈值分割图像,经实验验证该改进算法更加准确和有效,平均正确检测率达96.9125%,平均Kappa系数达0.6075。(6)总结本文的工作,展望未来SAR图像变化检测技术的发展趋势。
[Abstract]:Change detection, which is an important branch of remote sensing image research field, is to analyze and determine the feature and process of ground object change from the same area and different phase image, which is compared with visible light imaging. It has the characteristics of multi-polarization, all-weather, all-day and certain penetration, so it is suitable for the research of change detection. However, because of the speckle noise in SAR images, SAR images are different from other remote sensing images, and a unique change detection method is formed. Aiming at improving the performance of the change detection method of SAR image, this paper focuses on the change detection technology of SAR image. The main work can be summarized as follows: (1) introduce the principle of SAR correlation and the theory of change detection. The general flow and basic methods of SAR image change detection are summarized, and the mechanism, model and statistical characteristics of speckle noise are analyzed. According to the results of SAR image change detection, the criterion and basis of accuracy evaluation are given. (2) the principle of Leehnian Frost and Gamma Map filtering algorithms is introduced in detail, the advantages, disadvantages and applicable conditions of different filtering algorithms are analyzed, and the experimental results are compared. Frsot filter is selected as the filtering algorithm in this paper. (3) the difference method, the ratio method and the logarithmic ratio method are compared and analyzed to construct the difference image. On this basis, an enhanced logarithmic ratio method is proposed to construct the difference image. In order to improve the separability of the changed and unchanged regions in the image. (4) an automatic threshold selection algorithm based on image segmentation is applied to the differential image constructed by the enhanced logarithmic ratio method. Based on the automatic threshold selection algorithm of Bayesian decision, Ki optimal threshold selection algorithm and EM iterative threshold selection algorithm are applied to the difference image constructed by enhanced logarithmic ratio method. This paper analyzes the advantages and disadvantages of the traditional algorithms and their limitations. In view of the shortcomings of the above algorithms, the image segmentation algorithm based on the minimum Tsallis cross-entropy is applied to the change detection of SAR images for the first time. The average correct detection rate is 96.1916 and the average Kappa coefficient is 0.58, which improves the performance of change detection. (5) an improved algorithm based on minimum Tsallis cross-entropy threshold selection is proposed to make full use of the spatial relationship between each pixel and the surrounding pixel in the image. This paper compensates a single threshold selected based on minimum Tsallis cross entropy, and uses the obtained dynamic threshold to segment the image. Experiments show that the improved algorithm is more accurate and effective, with an average correct detection rate of 96.9125 and an average Kappa coefficient of 0.6075. (6) the work of this paper is summarized. The development trend of SAR image change detection technology in the future is prospected.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【相似文献】