当前,经典的2维遥感变化检测方法不再适用于地质灾害引起的地表3维变化。此外,一般植被提取依赖近红外波段,而近红外影像未必能获取。因此,针对2010年和2011年汶川映秀地区的可见光航空立体像对,提出了一种检测植被和3维地形变化的方法。首先,生成研究区两个不同时期的DEM和DOM并完成配准,再利用CIE Lab色彩空间和Otsu分割算法对可见光DOM进行植被变化检测。然后,基于概率统计理论对新旧时期差分DEM提出了3维变化检测的自适应阈值确定方法,在高概率置信域条件下提取出地质灾害的高危区域。最后,对高危区域运用离散化积分方法估算出土方量的3维变化量。实验结果表明了所提出方法的可行性,有效性和实用性。本文不仅将常规遥感的2维变化检测升华至3维空间,而且对变化的3维地形进行了定量估计,可应用于地质灾害的遥感动态监测和评估。

The classical methods of two dimensional remote change detection are no longer working effectively in three dimensional change scenes, such as geographical disaster areas. Vegetation information extraction also generally requires near-infrared bands. Visible light images from 2010 and 2011 in Yingxiu are selected and vegetation and three dimensional terrain change detection methods are presented under conditions without near-infrared bands. The Digital Elevation Model(DEM)and Digital Ortho Map(DOM)in two periods are generated by DPGrid. Then, vegetation changes in DOM are detected by the CIE Lab and Otsu algorithms. This study proposes an adaptive threshold method of three dimensional change detection based on the theory of probability and statistics. After censored samples are selected, the threshold range of elevation change is obtained by the 3σ rule of Gaussian distribution. High risk areas of geological disasters are extracted under the condition of high probability confidence regions. Finally, the corresponding earthwork change quantities are estimated by discretized integral. Vegetation changes in two periods are successfully detected, and change areas are located on the edge of the canyon in floodplain. This study determines the elevation change threshold range of three dimensional change detection by calculation. The threshold values of elevation decrease and increase are 2.73 m and 2.43 m, respectively. Landslide high risk areas and two debris flow accumulation areas located along Minjiang River are successfully detected. The earthwork quantitative change of 10 landslide high risk areas and two debris flow accumulation areas are estimated. Results show that the proposed method is effective, feasible, and practical. This study not only elevates conventional two dimensional change detection into three dimensional space but also quantitatively estimates three dimensional terrain changes. Therefore, the proposed method can be applied to monitor dynamic remote sensing and evaluate geological disasters. Even without near-infrared band, the proposed method can still successfully detect vegetation changes. In this study, statistical methods are used to determine the elevation change threshold. In allusion to landslide disaster, elevation-significant reduction risk areas are extracted. The distribution of risk areas is consistent with other research results. With discretized integral calculation, this study explores the earthwork quantity estimation method based on DEM. This method breaks the assumption that the surface is continuous and gradual, and can quickly estimate earthwork quantity by DEM. This method also incurs some errors caused by limited data source. The quantitative analysis for the uncertainty of the error is the main research target in the future. The two DEMs are generated separately in two periods; thus, they have their own errors in three dimensional change detection. Errors may also be amplified in error propagation. In future studies, the geographic information system data from previous periods will be used as control information to generate a DEM for a new period. Doing so would reduce relative errors. A domain-specific model will be also used to analyze and evaluate geological disasters.