Automatic Identification Of Landslide Based On GF-1 Satellite Image

The occurrence of landslide has brought a great disaster to the production and life of human beings.The forecast of landslide is still a complex problem at present.Because most landslides occur in the mountainous area with the complicated terrain and inconvenient transportation,that is what make it much more difficult to investigate landslides on site and to carry out the post-disaster rescue and reconstruction work quickly.In this paper,fast identification of landslides in Shuicheng county of Guizhou province was performed by the combination of remote sensing technology and computer recognition,which was depended on GF-1 image to remedy disadvantages mentioned above.Putting forward to using cloud transform and Fast Generalized Fuzzy C-means(referred to as FGFCM)algorithm,landslide recognition model is established based on ENVI,ArcGIS,eCognition and Matlab software.Finally,the identification results are verified.The main achievements and conclusions of this paper are as follows:(1)Through identifing the landslide in Shuicheng county,kinds of information is clearly expressed.Fistly,the size,nature and distribution characteristics of the landslide can be seen from the image.Secondly,it is easier to understand the impact of landslides on local people's land and roads,Thirdly,it is valuable to prepare for disaster prevention and mitigation during the rainy season and to reduce damage caused by landslides.The method can also be applied to other regions.(2)Before the fast identification of landslide,some ground objects with obvious spectral characteristics and easy to be misidentified as landslide are removed,which plays an important role in the accuracy of landslide identification.When extracting water system,roads,buildings,vegetation and other landslide disaster bearing bodies,selection factors and appropriate parameters are needed to extract each type of ground object.Only by repetitious comparing and analyzing the segmentation results can the optimal segmentation parameters and factors be suitable for different land classes to be selected.Four optimal segmentation parameters and factors are determined during the process of repetitious experiments,and then these parameters were applied to extract the landslide bearing body in 27 segmentation regions.The extraction accuracy reached more than 88%,and the results were satisfactory.(3)This paper used cloud transform and FGFCM algorithm to identify the landslide in Shuicheng county.It mainly depends on the pixel brightness value of the same ground object to fluctuate in a small range,and then used cloud transformation to figure out the problem of determining the initial central value,getting the number of classifications in FGFCM algorithm and reducing the human intervention.Then,on the basis of eliminating the water system,roads,buildings,vegetation and other landslide disaster bearing bodies,the landslide in Shuicheng county was identified by FGFCM algorithm,and the obtained results were validated.Among them,the kappa coefficient reached 75.18% and the overall accuracy reached 88.07%,which can meet the application requirements of actual production.(4)Using the cloud transform and FGFCM combination,the whole process and the finalresult are clear and organized accompaning by certain charts with the aid of ENVI,ArcGIS,eCognition and Matlab software,with a certain chart to explain the whole process.The results of landslide identification are also in line with the actual needs only a small amount of landslide knowledge,as long as skilled software operation skills,landslide identification analysis can be carried out.