Extraction And Research On Building Information From High-resolution Remote Sensing Image For Seismic Risk Assessment

Earthquake disasters are unpredictable,which seriously threaten the safety of human life and property,as well as economic development and social stability.Limited by the current level of science and technology,human beings cannot accurately predict the occurrence of short-term earthquakes.The large-scale earthquake risk assessment before the earthquake is not only helpful for the government departments to develop urban planning and emergency management plans,to minimize the risk of earthquake disasters,but also to improve the people's awareness of the risk of potential disasters.The building data is the core element of the seismic risk assessment.The traditional field survey method has the advantages of high accuracy and high reliability,but it also has the disadvantages of time-consuming and slow updating speed,which can not meet the needs of regional-scale seismic risk assessment.In recent years,with the rapid development of the economy and society,China is experiencing a large-scale and rapid urbanization process.The continuous expansion of city scale and the rapid and continuous change of buildings not only aggravate the potential seismic risk but also increase the difficulty of timely and accurate seismic risk assessment.Therefore,the study and establishment of a set of large-scale,high-precision,fast building data acquisition,and timely update methods can provide necessary data support for regionalscale seismic risk assessment;at the same time,considering the social vulnerability factors to build the earthquake scenario,to explore the vulnerability of buildings and the degree of social vulnerability to the earthquake scenario,and to deploy targeted.The work of earthquake emergency preparation has important application value and practical significance for minimizing the risk of the earthquake disaster.To solve the difficulty of building data acquisition and slow update in the current regional-scale seismic risk assessment,this paper first proposes two building segmentation models based on convolution neural network(CNN)to achieve highprecision,fast and automatic extraction of building contour.We built the building dataset of Urumqi from Google Earth high-resolution remote sensing image and the rural building dataset of Weinan from Unmanned Aerial Vehicle(UAV),ARC-Net model is applied to extract buildings with high precision and speed,and its practical application effect in earthquake risk assessment is verified.On this basis,the method of machine learning and data mining is adopted to solve the problem.According to the multi-source data such as street view image,road traffic,land use planning,statistical data and multi-temporal remote sensing image,the corresponding rules between building attribute information and building vulnerability are established;Taking Urumqi as the research area,the above calculation results are compared with the field survey to verify the accuracy and reliability of the methodology;the seismic risk assessment under different intensities are analyzed.Finally,taking the rural Weinan area as research area,the vulnerability of buildings to earthquakes,social vulnerability factors,and the impact of scenario earthquakes are explored.The main research work of this paper is as follows:(1)In this paper,the research status of seismic risk assessment theory and building information extraction methods are summarized.On this basis,aiming at seismic risk assessment,an automatic building extraction model of high-resolution remote sensing image is proposed based on deep learning neural network.The specific steps are as follows: the advantages and disadvantages of the existing convolutional neural network model are analyzed;then,a light-weight convolutional neural network model,USPP,which integrates the structure of encoder-decoder and spatial pyramid pooling module,is proposed.To verify the accuracy and efficiency of the USPP model,two public building datasets were compared: Massachusetts dataset and INRIA dataset.The results show that compared with other widely-used CNN models(including Seg Net,FCN,Unet,Tiramisu and FRRN),USPP model improves both qualitative and quantitative results.Compared with U-Net,the overall accuracy of USPP model in the two datasets increased by 1.0%(0.913 vs.0.904)and 3.6%(0.909 vs.0.877),while the training time of the model increased by 3.6% and 1.0%,respectively.Results show that the proposed USPP model can be used to segment buildings,with small classification errors and clear boundaries.(2)Although the result of building segmentation that USPP model achieved is satisfied,it brings additional computing resources consumption.The computing efficiency can not meet the needs of large-scale acquisition of building data in regionalscale seismic risk assessment.To improve the efficiency of deep learning,a novel convolutional neural network model,ARC-Net,is proposed.The main purpose of the ARC-Net model is to reduce the model parameters and to speed up the deep learning calculation,the core RBAC module of ARC-Net network is designed by combining the depth-wise separable convolution and asymmetric convolution;at the same time,the dilated convolution and the atrous spatial pyramid pooling module are used to expand the receptive field and to achieve better semantic segmentation results.Experiments on two building datasets,INRIA and WHU dataset,show that the proposed ARC-Net model has better performance than the other state-of-the-art deep learning models,with higher accuracy and less calculation cost.The model successfully extracts buildings with low classification error and shape boundary.At the same time,based on the Google Earth image and UAV image,the rural building datasets of Urumqi and Weinan are established respectively,and the ARC-Net model is applied to the dataset for automatic building segmentation.The experimental results show that the overall accuracy of building extraction in Urumqi is 0.899,and the precision is 0.848;the overall accuracy of building extraction in the rural Weinan area is 0.929,and the precision is 0.876.The outline of the building can be well segmented.Results show that the ARC-Net model can achieve higher accuracy of building extraction and calculation efficiency,and the results of building extraction in rural areas are better than that in the urban area.(3)Based on building extraction in the last step,to quickly and accurately obtain the building attribute information and to carry out the regional-scale seismic risk assessment,this paper proposes a comprehensive method based on multi-source data fusion,machine learning,data mining,and geographic information science,and takes Urumqi as an example.First,some areas of Tianshan District are selected to carry out the detailed field survey of buildings and to establish the field survey database of buildings;the EMS-98 building vulnerability classification standard and two kinds of data mining methods,support vector machine and association rule learning,are utilized to establish the corresponding relationship of building information and its seismic vulnerability;then,based on street view image,traffic data,and multi-temporal remote sensing image,this paper divides the residential block in Urumqi,and applies the corresponding rules of building vulnerability established in the last step to the Urumqi database,with an overall accuracy of 79.7%;finally,the seismic risk assessment model is used to analyze the distribution of seismic risk in Urumqi under different seismic intensities.When the seismic intensity is VIII and the soil effect is not considered,most buildings in Urumqi will suffer "slightly" damage.The results show that the method of building information extraction and seismic vulnerability assessment based on multitechnology and multi-source data fusion established in this paper can be used to assess the regional-scale seismic risk at a far lower cost than the traditional methods(such as field survey),and achieve ideal results,which has important application value.(4)Based on the comprehensive seismic risk assessment and analysis of the vulnerability,risk,and social vulnerability of scenario earthquake,taking rural Weinan area as an example.Firstly,three villages with different topographical characteristics in Weinan area are investigated.Then,based on the field survey data,the RISK-UE vulnerability index method is applied to quantitatively analyze the damage degree of buildings in three villages under different seismic intensities.Finally,considering the social vulnerability factors,the scenario-based earthquakes are built with the other earthquake impact scenarios of the three villages,which provides a good supplement for the earthquake risk assessment in Weinan area.All calculation results are calculated and displayed in Arc GIS software.The average building vulnerability indexes of Helan Village,Zhaojia Village,and Dongyu Village are 0.69,0.70,and 0.76,respectively.Under the same earthquake intensity,the damage of Zhaojia Village is the highest while Helan Village is the lowest.When the Weinan earthquake with intensity above VII,Dongyu Village will suffer landslide and traffic isolation.Residents will only rely on self-help and mutual help.