Building Seismic Damage Information Extraction Based On Full Polarization SAR

China is a country with frequent earthquakes.Destructive earthquakes often bring huge casualties and property losses to China.Today,earthquake prediction is still a worldwide problem.Rapid emergency rescue measures are the main channel to reduce disaster losses.With the development of science and technology,remote sensing technology is gradually applied to emergency rescue work,especially SAR remote sensing technology,with the advantages of all-weather and all-time observation,which plays an important role in emergency rescue Huge impact.In this paper,the building damage information in Yushu area is extracted quickly and accurately by using the full Polarimetric SAR remote sensing data and geographic information POI data of Yushu earthquake area in 2010.In this paper,the research status of building seismic damage assessment using Polarimetric SAR data in recent 10 years is discussed.The main data types are multi-source remote sensing data,multi-temporal remote sensing data and single temporal remote sensing data.The main research methods can be divided into polarization based and texture based.Secondly,in order to explain the imaging principle of Polarimetric SAR,a special chapter mainly introduces the image theoretical basis of SAR.The Polarimetric decomposition model(Yamaguchi4,Pauli),gray level co-occurrence matrix decomposition model and Polarimetric characteristics involved in this paper are discussed.Then,this paper takes the full polarization SAR image of Yushu earthquake in 2010 as the research object,and divides it into two chapters based on texture information and combined polarization information,using the methods of machine science and neural network for experimental analysis.(1)The texture features of buildings calculated by variogram have high resolution.It is easy to extract collapsed and intact building areas by machine learning method,with the accuracy of 80.14% and 84.54% respectively.However,the premise of extraction is to draw the mask of building area manually,which is not conducive to the promotion and application of actual earthquake cases.(2)In order to make full use of the texture information and Polarimetric information of the full polarization SAR image,this paper introduces the convolution neural network(CNN)method to train the convolution neural network respectively for the intact buildings,collapsed buildings and background objects in Yushu area.Finally,the whole SAR image is cut into the trained convolution neural network and the combined result classification map is obtained.The accuracy of the whole extraction is 99.12%,and the whole process is completed by the computer independently,which is conducive to the popularization and application of actual earthquake cases.(3)POI data is incomplete in the western region,but it is available in big cities.Therefore,in the results,only part of POI data can be used for the final comprehensive evaluation of earthquake damage,which also shows that the seismic damage evaluation model combined with POI data in this paper is more generalized in urban earthquake.