基于NDVI变化检测的滑坡遥感精细识别

随着遥感技术的发展，高分辨率的卫星影像数据逐渐丰富，滑坡灾害的信息提取被进一步推进，当前滑坡灾害应急调查主要以目视解译和野外调查为主，费时费力，难以满足灾后救援的迫切需求。面向像元和面向对象的单时相滑坡遥感信息提取方法等存在着滑坡过识别、误识别的问题。因此，在此提出以滑坡前后多时相遥感影像为数据源的变化检测滑坡识别方法，首先根据归一化植被指数（NDVI）进行基于像元的变化检测确定滑坡预选区，再结合面向对象的几何规则完成滑坡的精细识别，这种基于变化检测和几何规则相结合的方法能有效排除道路、建筑、裸地等光谱特征与滑坡相似的非滑坡部分。以九寨沟滑坡为例，采用高分一号8 m分辨率多光谱相机2015年8月1日的影像（滑前）以及2017年8月16日的影像（滑后）作为数据源，进行滑坡识别实验。结果表明，和面向对象的单时相方法相比，基于变化检测和几何规则相结合的多时相方法滑坡提取的精度较高，制图精度高达88.80%，用户精度高达81.19%，都大幅超过面向对象单时相法的精度，漏分误差及错分误差分别下降23.22%和11.72%，可为有效组织滑坡灾后救援与重建工作提供可靠依据。

Landslide Identification Method based on NDVI Change Detection

With the development of the remote sensing technology， the high-resolution satellite data is gradually enriched， and the information extraction of landslide disaster is further promoted. The current emergency investigation of the landslide disaster mainly focuses on the visual interpretation and field investigation， which is time-consuming， laborious， and difficult to meet the urgent need of the rescue after disaster. The single-phase landslide information extraction methods by using remote sensing based on the pixel-oriented or object-oriented have problems of over-recognition or mis-recognition of landslides. Therefore， the multi-temporal landslide information extraction method is worth studying and is expected to achieve good results， especially through the notable NDVI change in landslide. First， multi-temporal remote sensing images before and after the landslide are used as the data source. The landslide pre-selection area is determined using the pixel-oriented NDVI change detection. Then， the object-oriented geometric rules are used to complete the fine identification of landslides. This method based on the combination of the change detection and geometric rules effectively eliminates non-landslide parts which are with the spectral characteristics similar to landslides， such as roads， buildings， and bare land. Taking Jiuzhaigou landslide as the study case， the Gaofen-1 multi-spectral images of August 1， 2015 （before Jiuzhaigou earthquake） and the images of August 16， 2017 （after the earthquake） are used as data sources to conduct landslide identification experiments. The experimental results show that the multi-phase method has high accuracy in landslide identification. Compared with the object-oriented single-phase method， the former method has a mapping accuracy of up to 88.80% and the user accuracy up to 81.19%， both of which greatly exceed the accuracy of the object-oriented single-phase method. Moreover， the omission error and the mis-classification error decreased by 23.22% and 11.72%， respectively. This method determines landslides through the change of NDVI and has high timeliness in landslide identification， which does not need to consider the restrictions of excessive topographic and geomorphic factors and can be applied to most areas. It is believed that our method can provide a reliable basis for the effective organization of rescue and reconstruction work after landslide disaster.