Object-oriented Identifying Landslides Using Remote Sensing And DEM Data

As a kind of geological disasters, landslides causes many casualties and considerable property damage every year. Given the grim situation of landslide disasters and the enormous loss which was caused by landslide, investigation to identify the specific distribution of landslides is the foundation of the further study of landslide. In Gansu Longnan in Wudu Section of BaiLongJiang River Basin, landslide is one of the geological disasters which occurs frequently. This paper put the identification of the landslide disaster in this region as the content of the research. This article using the ZY-3and Landsat8OLI images as the remote sensing data, and using the Object-oriented and multi-scale image segmentation method which can segment the ZY-3image, generate100,150the two scales images. Meanwhile the object which was segmented replacing the recognition unit which was based in the pixel, was used as the basic unit of landslide identification. Then selecting the spectral, shape, index as the remote sensing features, and select the height, slope, mean curvature, distance from the channel length as the terrain features, this19features established the set of features to identify landslide. Then using linear, polynomial and RBF kernel function these three support vector machines establish landslides recognition model in two scales. Subsequently, identifying the landslides in the study area,then do Analysis of Variance between features of the result. The result shows:1. different segmentation scales effect the identification of landslide, the overall scope of recognition is similar and there are some differences locally;2. using SVM with different kernel functions is different in recognition accuracy, the recognition accuracy of RBF kernel function is best;3. using Object-oriented segmentation technology and SVM, the accuracy of the study area overall70%;4. analysis of variance between features of the result shows the selected features of non-landslides and landslides have significant differences.