基于面向对象分类法的川藏铁路沿线大型滑坡遥感解译
Remote Sensing Interpretation of Large Landslides Along SichuanTibet Railway Based on Objectoriented Classification Method

摘要： 川藏铁路位于青藏高原中东部，其地形地貌和地质构造复杂，是我国大型滑坡发育最为密集的地区，严重影响了我国西南地区的人类生命财产安全和重大工程建设。在面上资料分析的基础上，重点以川藏铁路沿线茶树山滑坡、川藏公路102道班滑坡、八宿怒江滑坡、理塘乱石包高速远程滑坡为例，在ENVI51和eCognition软件平台上，利用高分辨率WorldView2以及Landsat遥感卫星数据，结合野外实地调查，采用基于面向对象分类法对滑坡的遥感信息进行分析研究。结果表明，采用面向对象分类法可以提取出关键信息和目标区域，再结合目视解译，能够得到滑坡的细部信息，提高遥感影像滑坡解译的成功率，特别是对川藏铁路沿线等地质条件复杂区域的滑坡调查工作有重要意义。最后，结合灰度共生矩阵(GLCM)和归一化植被指数(NDVI)对古滑坡和新生滑坡的识别进行探讨。基于灰度共生矩阵和植被指数提出了滑坡遥感信息量判别(GVI)模型，并构建模型的质量函数IGVI；统计样本的结果显示古滑坡的IGVI值明显低于新生滑坡，表明本研究提出的GVI模型可以为识别古滑坡和新生滑坡提供依据。
Abstract: The SichuanTibet Railway is located in the Middle and East of the Tibetan Plateau, where the landforms and geological structures are very complicated. The large scale landslides are densely developed in this area, and they have caused serious human life casualties and losses of major projects. Based on the zonal remote sensing image data, taking the example of Chashushan landslide, 102 Daoban landslide, Nujiang landslide and Luanshibao landslide along the SichuanTibet Railway, this study takes the method of objectoriented classification to analyze the remote sensing information of landslides by use of high resolution WorldView2 and Landsat TM remote sensing image data on the ENVI51 and eCognition software platforms. The result shows that the key information and target area can be extracted by objectoriented classification method, and combined with visual interpretation the details of landslides can be obtained, which can improve the success rate of landslide interpretation. It is of great significance to the landslide survey in the complicated geological conditions along SichuanTibet Railway. Finally, this study combined the greylevel cooccurrence matrix (GLCM) and normalized difference vegetation index (NDVI) to discuss the identification of the ancient landslide and the reactive landslide, and on this basis supposed to put forward the GVI model and constructed the mass function IGVI of GVI model. The results of statistical samples show that the IGVI value of the ancient landslide is lower than that of the reactive landslide, indicating that the GVI model proposed in this paper can provide a basis for identifying the ancient landslide and the reactive landslide