多时相SAR图像多尺度变化检测

发布时间：2018-12-12 12:59

【摘要】：尺度特性是遥感图像所具有的重要特征,不同尺度下的SAR图像会表现地物不同的尺度信息。针对如何有效利用尺度信息实现地物变化信息的完整描述,进而提高SAR图像变化检测的精度和稳健性这一问题,论文围绕SAR图像的多尺度特征展开分析,对SAR图像自动配准及变化检测技术进行了研究,主要完成了以下几个方面的工作:(1)针对尺度不变特征变换(SIFT)在SAR图像中同名特征点匹配效率不高且准确性不够的不足,采用图像分块思想改进了基于SIFT特征的SAR图像配准算法。首先,对已粗配准的SAR图像进行分块,利用SIFT分别提取对应子块的特征点,形成全图均匀分布且准确度较高的候选特征点。然后,利用随机抽样一致性(RANSAC)算法筛选出稳健特征点,进而完成后续的配准。实验结果表明,该方法不仅可以显著缩短特征匹配的时间,而且能有效地提高配准精度。(2)针对多数变化检测方法难以准确鉴别变化区域边缘像素的问题,提出了一种结合结构相似性的多尺度SAR图像变化检测方法。其中,高斯多尺度变换用来构造像素的特征描述子,结构相似度则用来自适应地选取最优的尺度因子。通过对该尺度因子下的特征描述子进行模糊C均值分类,最终实现变化像素与未变化像素的分离。实验结果表明,该方法显著地提高了变化区域边缘的定位性能和变化检测的精度。(3)为提高变化检测的精细程度,提出了一种基于马尔可夫随机场融合的多尺度变化检测方法。该方法提出了一种小波独立重构方式,获取不同尺度下保持良好细节信息的图像集,进行均值循环迭代分割后,采用马尔可夫随机场进行变化检测结果的融合,得到最终的变化检测结果。独立重构充分利用了小波的高频系数,因此较好地刻画图像的细节;而结合马尔可夫随机场融合,综合考虑了SAR图像的空间邻域信息和不同尺度信息,进一步使得变化检测的结果更为准确。
[Abstract]:Scale characteristic is an important feature of remote sensing image. SAR images with different scales will represent different scale information of ground objects. In order to improve the accuracy and robustness of SAR image change detection, the paper analyzes the multi-scale features of SAR image, aiming at how to realize the complete description of ground object change information by using scale information effectively, and then improve the accuracy and robustness of SAR image change detection. The automatic registration and change detection of SAR images are studied. The main works are as follows: (1) aiming at the lack of efficiency and inaccuracy of scale-invariant feature transform (SIFT) matching in SAR images with the same name, The algorithm of SAR image registration based on SIFT features is improved by using the idea of image partitioning. Firstly, the coarse registered SAR images are divided into blocks, and the feature points of the corresponding sub-blocks are extracted by using SIFT to form candidate feature points with uniform distribution and high accuracy of the whole map. Then, the robust feature points are screened by random sampling consistent (RANSAC) algorithm, and the subsequent registration is completed. The experimental results show that this method can not only shorten the time of feature matching, but also improve the registration accuracy effectively. (2) aiming at the problem that most of the change detection methods are difficult to identify the edge pixels of the changing region accurately, A multi-scale SAR image change detection method combining structural similarity is proposed. Among them, Gao Si multi-scale transformation is used to construct pixel feature descriptor, and structural similarity is used to adaptively select the optimal scale factor. By classifying the feature descriptors in the scale factor with fuzzy C-means, the separation of the variable pixels from the unaltered pixels is finally realized. The experimental results show that this method can significantly improve the location performance and accuracy of change detection. (3) in order to improve the precision of change detection, A multi-scale change detection method based on Markov random field fusion is proposed. In this method, a wavelet independent reconstruction method is proposed to obtain the image sets with good detail information at different scales. After iterative segmentation of the mean value loop, the Markov random field is used to fuse the change detection results. The final change detection results are obtained. The independent reconstruction makes full use of the high frequency coefficients of wavelet, so it can depict the details of the image. Combined with Markov random field fusion, the spatial neighborhood information and different scale information of SAR images are considered synthetically, which makes the change detection results more accurate.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】