Research On Building Earthquake Damage Extraction Based On UAV Oblique Images

UAV remote sensing plays an important role in earthquake damage assessment of buildings.It can quickly take photos of buildings after the earthquake,extract damage information from buildings,estimate the extent of damage to obtain disasters,and provide decision-making materials for national rescue.Traditional remote sensing can only take aerial photographs of the top of the damaged building,and it is difficult to obtain the side information of the damaged building.As a result,the generated orthophoto image lacks the information of side,height and spatial location of buildings,which affects the evaluation of the damage level of buildings.In recent years,the UAV tilt photography technology has taken photos of the top and side of buildings from multiple angles.The 3D model generated by the model not only reproduce the 3D scenes of earthquake disasters such as building damage,landslides and barrier lakes,but also reflect the details of the side and outer walls of buildings.Taking the 2017 Jiuzhaigou Ms7.0 earthquake as an example,this paper proposes a method for extracting the building damage information from the three-building images with a resolution of 10 cm after Jiuzhaigou earthquake.The main research contents are as follows:(1)The three-dimensional model generated by the UAV oblique photography technology can better display the details of earthquake damage on the side and top of buildings.However,it is difficult to directly extract the earthquake damage information based on the oblique image due to the high latitude characteristics of the image,and the two-dimensional texture image transformed by reducing the dimension often leads to the incompleteness and fragmentation of the earthquake damage information of buildings.To solve these problems,this paper proposes a method for scattering the three-dimensional model,separating texture image from trianmulated irregular network,and directly obtaining the complete texture image after Jiuzhaigou earthquake.The optimal texture image is selected by using the tile coordinate range of pyramid model,the naming rules of tile and the spatial position of building monomer.After the optimal segmentation scale of building exterior wall in texture image is determined by using weighted mean variance method,this paper adopts the object-oriented method to extract the information of building exterior wall and wall skin shedding.Through the analysis of earthquake damage characteristics of these buildings,the damage level of single building is determined.The results show that the method obtains the complete four side earthquake damage texture images of buildings,and extracts the information of the external wall,crack and wall peeling area based on the texture image to determine the medium and serious damage levels of the single building.(2)Combining object-oriented multi-scale segmentation algorithm and deep learning convolution neural network training model,buildings in disaster areas can be extracted automatically and quickly.Object-oriented multi-scale segmentation can divide different objects into different objects.However,the description of features such as band,shape,location and texture is not complete,which reduces the efficiency and accuracy of recognition and classification of objects.Convolutional neural network can effectively recognize different objects through multiple feature training.In this paper,we firstly use visual interpretation to interpret the earthquake damage information of the top and side images of building groups,and select the top and side of buildings with earthquake damage information to construct training sample area.The sample size is 100*100 size pixels.The sample area can be divided into three categories: intact building surface image label 1,building surface image label 2 with damage information,and other objects and background image label 3.Then the convolution neural network VGG-16 model is used to train,calculate and classify the sample area.After the model training is successful,the multi-scale segmentation algorithm is used to segment the top and side images of buildings.The constructed model is used to extract the top and side earthquake damage information of the buildings,and the results are compared with the visual results.In this paper,the total area of the oblique image after Jiuzhaigou earthquake is about 0.3 square kilometers.The amount of buildings is small,which has a certain impact on the extraction results.And the essence of this model is the pyramid image constructed by the equally spaced tile segmentation,which increases the difficulty of extraction.