Landslide Recognition Based On High Resolution Remote Sensing Images In Heifangtai

Since 1968,loess landslide disasters continue to occur in Heifangtai area,which is known as the "modern landslide natural laboratory".The landslides threaten lives and properties,and cause significant damage to environment and natural resources.Detection of landslide is the foundation of landslide study,widely applied in landslide investigation and landslide mapping,hazard assessment,and can provide a scientific basis for disaster prevention and mitigation work.With the rapid development of science and technology,the performance of remote sensing platform and sensor has been significantly improved,and the time resolution of satellite imagery has been greatly shortened,and its spatial resolution has been increased to 0.31 m.Nowadays,high-resolution remote sensing images are widely used in weather,ocean,cadastral survey and so on.Remote sensing for landslide technology has developed from the original artificial visual interpretation to recognizing landslide by computer,which include pixel-based information extraction technology and object-oriented image analysis technology that can make full use of high-resolution remote sensing images to detect landslide.Heifangtai in Yongjing County,Gansu province was chosen as the study area.Remote sensing data sources are high-resolution remote sensing images GeoEye-1 and WorldView-2.We use object-oriented image analysis method and supervise classification method to recognize landslide.Major research work and achievements are as follows:(1)This paper introduces the detail process of landslide recognition using object-oriented segmentation and classification method,which include the nature of multi-scale segmentation,selection enough object attributes to get the optimal feature space,establishing the appropriate classification rules and the method to evaluate accuracy.(2)We choose Heifangtai area as the research region and the data that we use are GeoEye-1 data,which acquired on September 11,2013,and a 30 m resolution digital elevation model(DEM)to extract landslide information based on eCognition software.We utilize spectral,spatial,texture and neighborhood characteristic information to build the rule set,which is used to identify landslides,finally the results are evaluated for accuracy.At the same time,the maximum likelihood supervision classification method is used to detect landslide.At last,after comparing the two methods,we find that the landslide recognition rate of the object-oriented image analysis method is 18% higher than that of the maximum likelihood classification method,and the landslide identified by the former is continuous and the boundary is smooth.Therefore,the object-oriented technology can better extract loess landslide hazard information while using high-resolution remote sensing images.(3)In this paper,we selected the remote sensing image of GeoEye-1 in September 2013 and WorldView-2 in November 2014 to carry out object-oriented change detection research in Heifangtai area.First,we adopt the object-oriented technology to discover landslide,in order to acquire the landslide map in two different images,then compare them to obtain the new landslide occurred during 2013.09 to 2014.11.Contrast with artificial visual interpretation results,we find that the correct extraction rate of the landslide is 25.0%,and the accuracy of the landslide extraction is 64.5%.The main reason is that the old and new region of the Dangchuan No.7 landslide is partially covered,the features of the landslides in Moshigou are similar to the surrounding area' s,and the new vegetation grow in the new landslide of the Jiaojia Village.This test reflects the reliability and limitations of object-oriented change detection landslide identification methods.(4)There are many ways to obtain landslide terrain parameters.We take advantage of three-dimensional visualization analysis in ArcGIS,overlay Worldview-2 image and the high-precision DSM got in 2015 through UAV aerial,to calculate the Duangchuan No.3 landslide's topographic parameters,which include slope,aspect,slope height,slope length,area,etc.In fact,the method is convenient,meet the practical application requirements,and could provide more information about landslide.