Study On Classification Of Post-earthquake Ground Objects Using Post-earthquake Airborne LiDAR Point Cloud Data

China is one of the countries who affected heavily by the earthquake all over the world.After the earthquake,the disaster area accompanied by rain,coupled with complex terrain environment,so it is the urgent problem to get the accurate and timely information and image of disaster area for emergency.Light Detection And Ranging(LiDAR)is known as Laser Range Finder.Compared with 2-dimensional plane information from images of SAR(Synthetic Aperture Radar)and optical image,The original data obtained by Laser Range Finder is discrete points with spatial coordinate information,echo intensity and number of echoes,etc.We can obtain the DEM((Digital Elevation Model)and improve accuracy of building extracted by LiDAR point cloud without positioning solution.LiDAR system can get the vertical structure of the building form which help us to achieve quantitative analysis of the degree of damage to buildings.Meanwhile compared with optical remote sensing,the impact of sunshine and weather on LiDAR system is smaller,it can work at anytime day or night and under all weather conditions.Therefore LiDAR point cloud can play a role in earthquake emergency.However,the current research and application of seismic damage assessment based on airborne LiDAR point cloud are few.The building damage is the main causes of casualties and economic losses due to earthquake.Therefore,it is very urgent to carry out the study on identification of building damage due to earthauke based on airborne LiDAR point cloud.This thesis analyzes airborne LiDAR point cloud features of different objects,especially ones based on individual object and performs classification test accorrding to the optimal selection of features,finally we get the effective identification of post-earthquake undamaged building,damaged building and tree in the partial area of Port Au Prince,the capital of Haiti,hit by a large earthquake with Ms7.0 in 2010.The focus of this paper is on the 4 key aspects as follows:(1)Point eluding data filtering methods are discussed,then Morphological Filter,Slope Filter and Cloth Simulation Filter are selected and compared to filter the LiDAR point clud on the study area.(2)Based on the elevation information of point cloud in the study area,the eCongonition tool is used to divide the individual objects through human-computer interaction.(3)The point clud features of individual objects for undamaged building,damaged building and tree are analyzed as a key research aspect.The author proposes for the first time the index of the ratio of cloud echoe number(R),which shows well effective in the extraction of trees from point clud of LiDAR.Further more,the indices of echo intensity,the nearest neighbor height difference,the normal vector angle and the slope are introduced,combineing with REN,for feature analyses of individual objects identified through the object-oriented region segmentation method(4)Classification of objects in study area are performed with LiDAR point clud.Three classification methods such as BP(Back-Propagation)neural network method,K-Nearest Neighbor Classification and Decision Tree induction are used to extract damaged buildings based on Matlab.Accuracy analysis shows that the classification accuracy of the three classification methods are relatively high,no less than 90%.It shows that the features and methods selected in the study are suitable to extract post-earthquake damage objects effectively,which implicates practical significance to enhance the ability of earthquake emergency.