【作者】 兰远鸽；

【导师】 余旭初；

【作者基本信息】 解放军信息工程大学 ， 摄影测量与遥感， 2010， 硕士

【摘要】 遥感影像变化检测是利用不同时段的遥感影像对感兴趣的地物、目标的变化情况进行辨别分析的过程。基于SAR影像的变化检测是遥感影像变化检测的重要分支。围绕其整个流程,本文对涉及的四个部分进行了深入研究,主要工作和创新点如下:1.设计了基于SAR影像的变化检测的技术流程,包括图像滤波、图像匹配、变化区域提取和精度评价。完成了一套工程化的变化检测系统。2.针对经典空域统计滤波器仅仅依靠滤波窗口大小来调节滤波能力的问题,引入并改进了相关邻域模型,更加合理地描述了SAR影像局部滤波窗口中的邻域信息,采用基于改进相关邻域模型的MAP滤波器,在抑制斑点噪声的同时很好的保留影像中的边缘信息。3.针对像素级变化检测中构造差异影像的问题,改进了差异影像的构造方法,采用差值融合比值的方法使差异影像中的差异信息更加丰富。对于从差异影像中提取变化区域的问题,引入了Bayes决策理论,在基于高斯分布的假设下,以非监督的方式找出最优变化阈值,实现了阈值选取的自动化。对于传统阈值方法没有考虑差异影像中相邻像元之间关系的情况,引入了马尔可夫模型进行了描述,改进了Bayes提取的方法。有效减少了检测结果中孤立像素的存在,提高了变化检测的精度。4.深入分析了像素级变化检测的不足,提出了面向对象的变化检测方法,引入了Mean Shift算法。Mean Shift算法是一种有效的非参数分割算法,其在基于色域-空间域的基础上,通过直接估计特征空间概率密度函数的局部最值获取未知类别的密度模式,并确定模式的未知,然后使之聚类到和这个模式有关的类别中去。实现影像的分割。这种面向对象的分割方法能够迅速、准确的划分出同质区域。为进一步的变化检测提供重要的依据。论文中所提出的算法都利用SAR影像进行了实验,并取得了比较满意的实验效果,可靠性和有效性都优于现有的经典算法。更多还原

【Abstract】 Remote sensing image change detection is the distinguishing and analyzing process of interested ground object or target changes, the change detection based on SAR image is an important branch of remote sensing image change detection. Around its entire process, this paper involves four parts and research depthly, the main point of the work and innovation are as follows:1.Designed the overall process of change detection based on SAR image, including image filtering, image matching, changes in regional extraction and accuracy evaluation.Also Completed a set of process-oriented change detection system.2.In order to solve the problem of the classical statistical filters relying solely on the window’s size of filtering to adjust the filtering capacity in the spatial, I introduce and improve the relevant neighborhood model, descripted a more reasonable of the neighborhood information SAR image in the local filtering window , using the relevant neighborhood model based on improved The MAP filter in suppressing speckle noise at the same time retain a good image of the edge information.3.The problem of constructing differences image in the pixel-level change detection, I improve the method of constructing differences image, using the difference integration ratio make the differences information much richer in the difference image. And then how to obtain the change region from the difference image.This problem can be solved using the Bayes decision theory, at the assumption that the Gaussian distribution, finding the optimal change use non-supervised means,achieving the threshold selection automation. For the traditional threshold method does not take into account differences in image context information, the introduction of the Markov model described in the context of information, effectively reducing the test results in the presence of isolated pixels and improve the change detection accuracy.4.Analysis of the lack of pixel-level change detection and the object-oriented change detection method is proposed, the introduction of Mean Shift Algorithm, Mean Shift algorithm is an effective non-parametric segmentation algorithm, and on the basis of the color gamut-spatial through direclyt estimated feature space probability density function of the local minimum value of the density model to obtain the unknown category, and determine the mode of the unknown, and then make it to the cluster, and this model to the relevant categories,achieving image segmentationThis object-oriented segmentation method to quickly and accurately classified into homogeneous regions. Providing an important basis for change detection in the further.The proposed algorithms are applied to SAR images , and obtained satisfactory experimental results, the reliability and validity are superior to the existing classical algorithm.更多还原