【作者】 张同刚；

【导师】 岑敏仪；

【作者基本信息】 西南交通大学 ， 大地测量学与测量工程， 2006， 博士

【摘要】 准确的地表形变信息是进行泥石流等自然灾害的预测、预报和防治等工作的基础数据。多时相遥感影像及其DEM产品是获取地表形变信息的主要数据来源。研究无控制DEM差异探测技术，利用多时相DEM监测地表变化，既可减少野外建立和维护控制点的工作，节省经费，又可提高地表变化监测的自动化程度，满足对自然灾害事件的快速响应要求，而且可将其应用范围拓展到过于危险、不易到达和缺乏明显地貌特征等难以建立控制点的地区。另外，该技术还能够利用由于年代久远而无法建立有效地面控制的灾前数据，充分发挥历史资料的价值。从表面上看，无控制DEM差异探测技术是利用表面匹配过程来替代建立控制点的过程，实际上在没有探测到表面差异时是不可能获得精确的表面匹配的，而精确的匹配结果是正确识别表面差异的前提。因此，表面匹配与表面差异探测是一个问题的两个方面，它们相互影响，使问题变得复杂困难。目前的研究成果尚不能够完全满足泥石流灾害地区应用的要求。本文以成都-昆明铁路线的普歪沟泥石流灾害地区为对象，研究自动探测大面积地表变形的无控制DEM匹配算法。选取适合DEM匹配的算法是研究工作的前提。通过试验分析现有的2种代表性算法，最小高差算法(LZD)和最近点迭代算法(ICP)，证实了LZD算法的综合性能指标更适合DEM匹配。接着提出建立DEM表面点对应关系的法向对应准则，与LZD组合，使得LZD算法在拉入范围、迭代收敛性等方面的性能得到了全面的改善。要探测DEM差异，先要判断观测量是否具有粗差可发现和可定位能力。运用误差与可靠性理论中的粗差可发现和可定位判断矩阵，证明了LZD算法判断矩阵中零列向量的个数等于LZD算法系数阵列秩亏数，试验验证了LZD算法匹配方程具备粗差发现和定位能力，为DEM差异探测算法提供了理论支持。根据泥石流地区山脊附近区域受泥石流影响小的特点，提出了基于广义控制点的分块区域匹配算法(MSPM)。它借助地性线提取算法来确定广义控制点。试验结果显示使用M-LZD的MSPM算法非常适合泥石流灾害地区存在大面积地表面变形的DEM匹配差异探测。现有的差异探测算法忽略了表面变形与粗差之间的区别，将表面变形作为粗差处理，仅考虑它们的数值大小，没考虑它们的相互联系。因此，顾及地表变形的特点，结合高崩溃污染率的截尾最小二乘估计(LTS)，提出基于差分模型的差异探测算法(DM-LZD)，进一步提高了在大面积变形情况下的DEM表面差异探测能力和匹配精度。模拟试验结果显示DM-LZD算法能够探测的变形面积超过50％，优于现有的M-LZD和LMS-LZD算法。在多时相DEM上提取向同大小的DEM子块，通过匹配DEM子块来获得整体DEM的转换参数初值，使得DM-LZD算法能够处理坐标系统相互独立的多时相DEM。利用普歪沟1957DEM和1987DEM的实际数据，DM-LZD算法定量测定了普歪沟30年来地表侵蚀和堆积的土方量，探测到的地表形变面积约为匹配DEM流域面积的58.6％，为今后普歪沟泥石流灾害的定量监测和评估奠定了基础。更多还原

【Abstract】 Precise Earth’s surface deformation is the foundational data for assessing, managing and predicting the natural disasters, such as debris-flows. The multi-temporal images and DEM are main data source for obtaining the deformation information. As to monitoring terrain changes with multi-temporal DEMs, researching DEM deformation detecting without ground control points (GCPs) will bring many benefits. Firstly of all, it would be helpful to reduce the field works of constructing and maintenance of GCPs and save money. Moreover, it would enhance the automatic degree for monitoring terrain changes, and then it would better satisfy the request of rapid respondence to natural disasters. Furthermore, it can make full use of those quite old pre-disaster data, although it is impossible to setup valid GCPs for them.Apparently, the technique of detecting DEM deformations without GCPs employs surface matching to replace GCPs. In fact, both surfaces could not be matched accurately before their deformations are identified correctly, while it is necessary for the precise matching. So surface matching and deformation detecting is two different aspects of one question, but their relationship and mutual interaction make this question be very complex. The existing researches cannot fully meet the requirement of real applications in debris-flow area. PUWAIGOU, a typical debris-flow valley along Chengdu-Kunming railway, southwest China, is adopted as experimental area to study the automatic DEM matching algorithm for detecting large deformed area.Selection of an appropriate DEM matching algorithm is the first task of this research. Two representative algorithms, least Z-difference (LZD) and iterative closest points (ICP) are compared by some simulated experiments. The testing results approve that LZD is more suitable for DEM matching according to its compositive performance. Then the normal direction corresponding criterion for pairing points on both DEMs is proposed. By integrating the novel criterion with LZD, the performance of improved LZD is enhanced greatly in pull-in range, convergence, etc.It should be confirmed prior to detecting deformation whether or not observations have detectability and locatability of gross errors. Using the judgment matrix of detectable and locatable of gross errors in surveying error and reliability theory, it is proved firstly that the number of zero column vectors in LZD’s judgment matrix equals to the rank defects of LZD’s coefficient matrix. And then, LZD’s matching equation having the detectability and locatability of gross errors is verified by a series of tests. These conclusions powerfully support the following researches on algorithms of detecting DEM deformations in theory.According to the character that the ridge region is rarely affected by the debris-flow activities, an algorithm called multiply surface patches matching (MSPM) based on generalized control points is reported. The generalized control points are selected automatically associated by the terrain features extraction algorithm. The experimental results show that MSPM, associated with M-LZD, is very practical for detecting large proportion of deformation in debris-flow area.The difference between gross errors and deformations is neglected in existing algorithms for matching DEM to detect deformations. The deformations are regarded as gross errors, therefore only their magnitude is considered and their relationship is not. So, considering the character of deformation, a novel algorithm called LZD using differential model (DM-LZD) is proposed associated with least trimmed squares estimator (LTS), which is a robust estimator with high break-down point. It enhances the deformation detecting ability and matching accuracy when large deformed area exists in matched DEM. The simulative experimental results illustrate that DM-LZD can detect over 50% deformation area, and is superior to both M-LZD and LMS-LZD.By matching the DEM blocks, which are of same size and are extracted from the multi-temporal DEMs, DM-LZD algorithm can be applied to the multi-temporal DEMs with independent coordinate system. Using the real 1957DEM and 1987DEM, the volume of soil erosion and debris deposition can be quantificationally detected by DM-LZD during the 30 years in PUWAIGOU. The detected deformed area is about 58.6% of valley area in matched DEM. Such information will form a foundation for quantificational monitoring and assessing PUWAIGOU debris-flow hazards in future.更多还原