Extraction Of Collapsed Buildings Using High Resolution Remote Sensing Imagery And LiDAR Data

In recent decades the earthquake occurred frequently. As one of the most destructive natural disasters, earthquake can cause a great deal of damage to human society.Based on the emergency relief after earthquake and the requirements to assess the information about the effect of the earthquake, and against the problems of complex terrain in the earthquake area, and various ways of buildings collapsed, and different characteristics showed in the remote sensing images, and the accuracy of only using remote sensing data sources to extract the collapsed buildings is usually not high, this study based on high resolution remote sensing images and LiDAR point cloud together, analyzed the rich texture and shape features of the images and the height information of the LiDAR points, through object-oriented multi-scale segmentation and support vector machine (SVM) classification method to extract the collapsed buildings, solved the problem of extracting collapsed buildings fastly and accurately after the earthquake, this paper has completed the following work:1) Combined the nDSM gained by processing the LiDAR point cloud, the terrain contour map gained by executing edge detection with Canny operator on the nDSM, with the high resolution images together, then carried out segmentation. It has been proved that after joining the LiDAR point cloud information in the high resolution images, the fault classification caused by the phenomenon of foreign bodies have the same spectral signature.2)This paper puts forward entropy of the gray level co-occurrence matrix as new homogeneity evaluation index after segmentation, combined this index with global Moran Ⅰ which is the index of heterogeneity, Constitute a new global optimal segmentation scale objective function, and experiments proved that the method is feasible.3) This paper determined the feature classification group used for extracting collapsed buildings in the experimental area. According to different land types in the experimental area, comprehensively utilizing the three characteristics of the high resolution remote sensing images:Spectrum, texture and shape features, and the height information of LiDAR point cloud, This paper presented a vegetation exctraction rules based on RGB images, and has been proved show high precision in trees extracting.4) This paper identified the process to extract collapsed buildings based on support vector machine (SVM) classification method, using radial basis kernel function as the kernel function of support vector machines (SVM), through the grid search and cross optimization to get the optimal kernel function parameter combination.5) Using support vector machine (SVM) classification method, based on the platform of Visual Studio 2010 and the computer language of C++, this paper completed the tests of extracting collapsed buildings, and evaluated the accuracy of the classification result.Based on the collapsed buildings extraction method proposed in this paper, we achieved a good accuracy: the producer’s accuracy is 92.22%, the user’s accuracy was 90.22%, and the Kappa coefficient is 0.857. The accuracy well proved the method is feasible and has a certain reference significance for carrying on similar-work, and it also provided theoretical the technical support for rapid assessment of collapsed buildings caused by the earthquake.