Information Extraction Of Landslides And Hazard Assessment Using Remotely Sensed Data

It is well-known that landslide is serious regardless of national boundaries and areas, the losses have been very great by landslides every year. When landslidehappens, the majority of people are caught off-guard, it has been a lot worse, landslide hazard assessment, the results of assessment can effective reflect the probability of landslide occurrence, it’s means something to a warning monitoring systemfor landslide.Hazard assessment using support vector machine model is little and hasnot to establish a complete evaluation system, paper choose mashu–tangdan inSichuan and Yunnan province as a study area,there ara lots of landslides in this studyarea. The purpose is to establish and verify a landslide hazard evaluation system usingsupport vector machine model. We choose logistic regression model to evaluate theaccuracy for quantitative evaluation,it is more convincing for the accuracy of theevaluation results.The paper also research band combination of SPOT5high-resolution image and image fusion of WorldView-2,the aim to improve theprecision of interpretation accuracy, then combine with field verification andcorrection to ensure the accuracy of the interpreted results,we analysis the relationshipbetween landslides and control factors and have a good grasp on the area of thelandslide distribution characteristics, the paper has made following understanding:(1)We statistics the mean, standard deviation, information entropy andcorrelation coefficient of each band of the SPOT5multispectral XS1, XS2, XS3, XS4XS5,(XS5is simulate the green band, in order to simulate true color image). In linewith the information entropy is maximum and the correlation coefficient between the band is minimum, the final choice is XS4, XS3, XS2,this combination is moreconducive to landslide extraction, the result can also be promoted in other areas.Wechoose three methods to fusion the data of WorldView-2panchromatic andmultispectral,and find that the principal component analysis is good for landslideextraction,because the method can make image more richer and clearer.(2)Landslide hazard assessment using Support vector machine model have animportant step in the hazard assessment: the choice of kernel function, radial basisfunction (RBF) does not appear wide of the mark in the process of calculation,because the parameter is less than kernel function,for example, the polynomial and Skernel function.the range is easy to control,RBF is to be recommended for nonlinearpartition.(3) The quantitative of evaluation factor is not clear, the authors propose a newquantitative methods:we statistic the elements of each evaluation factor landslide in arange containing the pixel area, using the area percentage of its to quantify theevaluation factors.(4)We have established landslide hazard evaluation system based on supportvector machine model and using different sample data for landslide hazardevaluation,(logistic regression model has used31%of the study area, support vectormachine model has used2.5%of the study area), support vector machine model ishigher than the logistic regression model to evaluate the accuracy and its were89.2%,80.8%,it is feasibility for landslide hazard evaluation system based on support vectormachine model,this system can be used by other gorge region.(5)The hazard assessment index reflects the possibility of the recurrence oflandslide area, areas of high index under the influence of the trigger factors such asrainfall, earthquakes, the possibility of the recurrence of landslide area is high andvice versa, as well as the evaluation of results intuitive reflect the high-risk arealocated along in the Jinsha river, the Xiaojiang river, this area is coincide with humanengineering activities, its confirm each other.