基于区域分割的多源、多时相卫星遥感影像联合匹配方法研究

发布时间：2018-01-24 02:43

本文关键词： 多源多时相卫星遥感影像影像匹配影像分割匹配传播预测位置修正轮廓线相关　出处：《武汉大学》2014年博士论文　论文类型：学位论文

【摘要】：随着数据获取手段的不断丰富,国内外在轨卫星数量的急剧增加,获得海量多源卫星数据已成为可能。单一卫星由于成像环境、重访周期等因素的限制,无法短时间内获得大范围覆盖的有效数据,所提供的信息己完全不能满足遥感对地观测数据广泛应用的迫切需求。而现代摄影测量逐渐发展进入一个多传感器、多光谱、多分辨率和多时相的新阶段。作为摄影测量处理的核心问题,影像匹配结果的好坏直接决定着最终产品的质量。由于目前受各种技术条件的限制及传统思维的束缚,针对多源、多时相卫星遥感影像的联合匹配技术尚不成熟,仍存在较多的问题需要解决。因此深入研究针对多源、多时相卫星影像的联合匹配方法,结合国内外卫星数据的特点,突破传统匹配思维的束缚,探索新的影像匹配方法与思路,对多源、多时相卫星遥感数据的联合摄影测量具有重要的意义。论文以不同传感器、不同时相差异的卫星遥感影像匹配作为研究目标,在摄影测量领域,创新性地将影像分割技术应用于匹配算法中,提出了一种基于区域分割的卫星影像联合匹配方法,设计了一套流程化的、行之有效的多源、多时相卫星遥感影像匹配方案。论文的主要工作如下： 1、结合现有影像分割技术,针对大数据量的卫星影像特点,本文提出了一种基于分裂-合并的影像并行分割方法。在分割方法中,本文构建了结合光谱特征、纹理特征、边界特征的合并代价度量准则,尽可能保证区域合并的正确性；设计了一种并行分割策略,并针对并行分割中出现的重叠区域分割结果不一致的问题,提出了基于层次树的分割结果无缝接边方法,保证了不同分割任务在重叠区域的分割结果一致性,避免了合并时造成的接边错位或断裂问题。在多机多核硬件条件下,通过本文提出的并行分割方法将影像划分为一系列局部连续、平滑的分割区域,用于后续匹配的空间约束；并利用基于游程编码的矢量化方法在分割结果中提取边缘轮廓线,用于后续轮廓线相关的匹配基元；另外,根据分割区域的相邻关系,将区域间的公共交叉点作为后续影像匹配的特征点,进行影像间的特征点匹配。因此在本文中,影像分割及其结果贯穿于算法的始终,用于指导影像的匹配,最终获得令人满意的匹配结果。 2、由于国产卫星数据的姿态轨道参数直接定位误差较大,同名点初始预测精度较低,导致了匹配约束条件如近似核线的失效问题。针对上述问题,本文提出了一种基于边缘轮廓线相关的方法,利用匹配获得的同名轮廓线在像方对姿态轨道参数进行修正,提高了同名点的初始预测精度,有效减小匹配搜索范围,增强匹配约束条件的可靠性,从而提高匹配的精度与稳定性。该方法首先利用基于支持向量机的云检测方法,对云覆盖的不可靠分割区域与轮廓线进行检测与剔除；其次构建了一种循环可变夹角链码,对轮廓线进行描述,通过链码间的相关性计算,确定局部最优的候选轮廓曲线段,避免了起始点不一致和链码顺序问题造成的匹配失效现象,并对实际应用中常见的同名轮廓线部分-部分对应情况具有较好的适用性；之后提出了一种基于HOGC (Histogram of Oriented Gradients based on Contour)的轮廓线匹配方法,在候选轮廓曲线段中实现了同名轮廓线的最终确定；最后通过同名轮廓线上关键点的对应关系,在像方对姿态轨道参数的同名点初始预测误差进行补偿。 3、通过对于现有匹配方法所存在问题的分析,结合多源、多时相卫星影像的特点,本文提出了一种全球SRTM (Shuttle Radar Topography Mission)辅助下的特征点匹配方法。通过对特征点的提取、匹配约束条件的确定、匹配策略的优化、误匹配的检测四个方面的研究,获得可靠且定位精度较高的特征同名点位,并作为种子点,为后续的匹配传播提供可靠的先验知识。 4、为获取更加密集的匹配结果,本文提出了一种基于分割区域约束的匹配传播方法。利用标记分割区域代替常用的三角网或多边形格网作为传播约束类型,并结合纹理与几何相似性,构建了一种联合距离、夹角、灰度分布的相似性测度-DANCC (An similarity measure integrates Distance, Angle, and Normalized Cross-Correlation),旨在增强纹理贫乏、重复区域、地形起伏较大区域的匹配正确性,提高匹配传播的正确率与可靠性。本文通过较为新颖的匹配思路,结合实际应用中所遇到的匹配问题,针对现有算法的不足,充分探索匹配困难区域的匹配可行性,提出了一种基于区域分割的多源、多时相卫星遥感影像联合匹配方法,将影像分割与影像匹配技术进行了深入的融合。为了证明本文方法的研究与应用价值,本文利用实际数据分别进行了多景大数据量的卫星影像匹配、Google Earth影像辅助的控制点自动匹配以及数字表面模型DSM (Digital Surface Model)自动生成的应用试验。通过试验结果证明了本文方法的有效性,为未来多源、多时相卫星遥感数据的联合摄影测量奠定了基础。值得一提的是,本文方法己成功应用于资源系列卫星、高分一号卫星等多个型号与工程项目中,在实际应用中匹配结果稳定可靠。
[Abstract]:With the means to obtain data continuously enriched, dramatic increase in the number of domestic and foreign satellites, get massive Multi-source Satellite data has become possible. The single satellite imaging environment, revisit cycle and other factors, not a short period of time to obtain valid data coverage, the information provided has been unable to meet the urgent needs of the earth the observation data are widely used. Modern photography gradually develops into a multi-sensor, multi spectral, multi-resolution and multi temporal new stage. As a core problem of photogrammetric processing, image matching results directly determines the quality of the final product. Due to the limitation of current technology, affected by various conditions and the traditional thinking for matching technology combined with multi-source, multi temporal satellite remote sensing image is not mature, there are still many problems need to be solved. So the in-depth research on The joint matching method of multi-source and multi temporal satellite images, combined with the characteristics of satellite data at home and abroad, breaks through the shackles of traditional matching thinking, explores new methods and ideas of image matching, and is of great significance for joint photogrammetry of multi-source and multi temporal satellite remote sensing data.
Based on different sensors, satellite remote sensing images at different phase differences, as the research object in the field of photogrammetry, innovativelinked image segmentation technique is applied to the matching algorithm, we propose a joint image matching method based on region segmentation, a streamlined design, effective multi-source, multi temporal the satellite remote sensing image matching scheme. The main work is as follows:
1, combined with the existing image segmentation technology, satellite images according to the characteristics of large amount of data, this paper proposes a parallel segmentation method based on Split Merge image segmentation. In the method, this paper constructs the combined spectral feature, texture feature, edge feature merging cost metric criteria, as far as possible to ensure the correctness of region merging; a parallel segmentation strategy is designed, and the parallel segmentation of overlapping segmentation results. The problem put forward the hierarchical tree segmentation results based on edge matching method, the different segmentation tasks in the overlapping area of the segmentation results of consistency, to avoid the merger caused by edge dislocation or fracture problems in the multi kernel hardware conditions, using the proposed parallel image segmentation method will be divided into a series of local continuity, regional segmentation, smoothing, for subsequent space constraints; And the use of vectorization method based on run length encoding to extract the edge contour in the segmentation results, relevant for subsequent contour matching; in addition, according to the relationship between adjacent region segmentation, feature points of the common intersection points between regions for subsequent image matching, feature matching between images. Therefore in this paper in the image segmentation and its results throughout the algorithm always used to match the image guidance, the final matching result is satisfactory.
2, because of the attitude and orbit parameters of domestic satellite data directly positioning error is large, the same point of initial prediction accuracy is low, leading to failure of the problem, constraint conditions such as approximate epipolar line. To solve the above problems, this paper presents a method of the edge contour line based on the use of the same name, obtain the contour lines in the image on the orbit and attitude parameters are modified to improve the prediction accuracy of the initial points, effectively reduce the search scope, enhance the reliability of matching constraints, thus improving the accuracy and stability of matching. This method uses the cloud detection method based on support vector machine, the cloud cover is not reliable to detect and eliminate segmentation and contour second line; building a circular alterable included angle chain code, description of the contour line, the correlation between the chain code calculation, to determine the local optimal candidate profile curve, To avoid the matching failure phenomenon caused by the starting point of disagreement and order of the chain code, and part of the same contour common application part corresponding situation has good applicability; after an algorithm based on HOGC (Histogram of Oriented Gradients based on Contour) matching method in the candidate contour line, contour curve the line is realized in the final name of contour line; finally through the corresponding relationship between the key points the same contour line, the compensation in the initial point of the same name as the parameters of attitude and orbit prediction error.
3, through the analysis, the problems existing matching methods combined with the characteristics of multi-source, multi temporal satellite images, this paper proposes a global SRTM (Shuttle Radar Topography Mission) feature point matching method. Assisted by extracting feature points, determine the matching constraints, matching strategy optimization research, four aspects of the detecting error, and obtain reliable and high positioning accuracy of the feature points, and as the seed point, provide prior knowledge for the subsequent matching propagation.
4, the matching results to get more intensive, this paper proposes a matching propagation segmentation method based on constraints. By labeling segmented regions instead of commonly used triangle or polygon grid as propagation constraint types, and combining the texture and geometric similarity, to construct a joint angle, distance, similarity measure -DANCC distribution (An similarity measure integrates Distance gray, Angle and, Normalized Cross-Correlation), is designed to enhance the texture is poor, the repeat region, the terrain matching accuracy of large area, improve the accuracy and reliability of matching propagation.
In this paper, a novel idea, combined with the matching problem encountered in the practical application, aiming at the shortcomings of existing algorithms, to fully explore the feasibility of matching matching difficult area, presents a multi region segmentation based matching method, combined with multi temporal satellite remote sensing image, image segmentation and image matching technology in-depth fusion in order to prove the method in this paper. The research and application of value, the actual data are satellite images view a large amount of data, the use of automatic matching and digital surface model DSM control point Google Earth images (Digital Surface Model) assisted the application test of automatic generation. Through the test results show the validity of this method for the future, multi-source, laid the foundation for joint photographic measurement of multitemporal satellite remote sensing data.
It is worth mentioning that this method has been successfully applied to many models and engineering projects such as resource series satellite, gou1 satellite and so on, and the matching results in practical applications are stable and reliable.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P237

【相似文献】