Building Damage Assessment From SAR Image Using Polarization And Statistical Model Texture Parameters

The collapse of buildings is a major factor in the casualties and economic losses of disasters,and the degree of building collapse is an important indicator for disaster assessment.Therefore,accurate building damage assessment provides an important basis for disaster relief and government decision-making.Synthetic aperture radar(SAR)has become an important mean for building damage assessment,due to its strong penetration and all-day/all-weather working capability.In this thesis,building damage assessment was studied from polarization and texture features of post-event SAR image.The main research contents are as follows:1)In order to make the model decomposition components closer to the scattering mechanism in urban areas,and improve the accuracy of collapsed building extraction,the Freeman decomposition was improved for the extraction of collapsed buildings in PolSAR image.The effectiveness was verified using RADARSAT-2 full-polarization data after Yushu earthquake.However,the intact biuildings with large oriention angele are easily misclassified into collapsed buildings.2)In order to distinguish large orientation intact buildings and collapsed buildings effectively,a statistical texture feature G0-para was proposed based on the G~0distribution texture parameter of of SAR data,which can be applied to single-/dual-/full-/compact polarization SAR.The receiver operating characteristic(ROC)curve and the area under ROC curve(AUC)was used to analyze the capability of G0-para for distinguish collapsed buildings from intact buildings,and analyze the effect of polarization models on the capability of distinguishing collapse and intact buildings.RADARSAT-2 and ALOS-1 data were used to verify the performance of G0-para.The results show that the G0-para of hybrid mode compact polarization(HYB-RC),VH/VV or HH/HV are similar to the PolSAR in distinguishing collapsed and intact buildings.Otherwise,for single-polarization SAR,VV also obtain better result.3)This thesis introduced the texture parameters of statistical model to enhance the distinguishability of collapsed and intact buildings,and proposed a building damage assessment method based on the PolSAR statistical model texture parameters.The texture parameters of Pol SAR statistical models were used to reflect the homogeneity of buildings,and were used to distinguish collapsed and intact buildings.Then,building damage was assessed based on the collapsed building rate in a block.Then,the building damage assessment results of different statistical models or different texture parameters estimators were analyzed by reference the interpretation result.Finally,the statistical model and texture parameter estimator suitable for building damage assessment were determined.The experiments were presented using the RADARSAT-2 and ALOS-1PolSAR data.Compared with other features,the results show that the texture features extracted from statistical model texture parameters avoid the comfusion between collapsed and intact buildings,and improve the accuracy of building damage assessment.The G~0 texture parameter estimation based on the second-order moment is most effective for PolSAR building damage assessment.There are two innovations in this paper:1)A statistical texture feature G0-para was developed based on the texture parameters of G~0 distribution,which can better distinguish between collapsed and intact buildings,and can be applied to single-/dual-/full-/compact polarization SAR data.2)A building damage assessment method based on the texture parameters of the PolSAR statistical model was proposed,which avoids the confusion between oriented intact buildings and collapsed buildings effectively,and assess the PolSAR building damage situation more accurately.