Research (Technology And Application) On SAR Multi-Feature-based Earthquake Building Damage Assessment

Synthetic Aperture Radar(SAR),with its all-weather and all-time observation capability,plays an important role in earthquake damage assessment.With the development of SAR imaging technology,its spatial resolution is improving gradually,and it has become one of the main data sources for earthquake damage information extraction.The method of extracting earthquake building damage based on multi-temporal SAR data can obtain more accurate results,but sometimes effective data are hard to obtain,which makes data processing more complicated.Therefore,building collapse assessment using only post-event SAR data has become an essential way in the process of earthquake emergency.There are various and rich features in SAR images,and each feature has a great difference in the ability to identify buildings.So far,there has not been a unified evaluation standard for combined application.With the development of the information extraction theory and technology,combining a variety of effective features can improve building recognition accuracy.However,when the selected effective feature dimension is too much,computing complexity of the algorithm will increase,making the extraction inefficiency.A feature fusion method is needed to reduce the operation efficiency while extraction accuracy are guaranteed at the same time.Aiming at the problem of low utilization and information redundancy of multiple features in identifying collapsed buildings with SAR images,this paper took a single ALOS-2 PALSAR full-polarization SAR image after the 2016 Ms7.0 Kumamoto earthquake in Japan as the research object.Supplemented with optical images and GIS data,the earthquake building damage information in affected areas was extracted.Specific research contents and progress were as follows:(1)Based on the imaging mechanism and imaging features of intact and collapsed buildings in high-resolution SAR images,the regularity analysis of SAR image features of quake-damaged buildings was carried out,and various texture and polarization features of damaged buildings in SAR images were analyzed.(2)Different features of texture and polarization information of image were extracted and discussed respectively.Based on texture and polarization features and combined with texture and polarization features,the extraction results and efficiency of earthquake damaged buildings are presented.The results show that the combination of texture and polarization characteristics can improve the information extraction of earthquake damaged buildings.Then,the results of building extraction based on K-Nearest Neighbor algorithm,Support Vector Machine algorithm and Random Forest algorithm are compared and analyzed.(3)The feature fusion based on PCA(Principal Component Analysis)and the characteristics optimization based on MRMR algorithm(Maximum Relevance Minimum Redundancy)were introduced.The two optimization algorithms are discussed on how they affect the result and extraction efficiency of the earthquake building damage extraction.,Results showed that MRMR,feature optimization combined with Random Forest classification has better effect in damage information extraction.The extraction accuracy is improved on the premise of ensuring operation efficiency.