Earthquake Damaged Buildings Extraction From SAR Images Using Change Detection Technology Based On Multiple Features

Earthquakes are one of the most serious natural disasters and can cause casualties and property loss.Earthquake damage assessment is an important step in such damage.Earthquakes can not only lead to the collapse of buildings and the direct loss of roads but also cause secondary disasters such as floods,fires,landslides and debris flows.Compared with optical remote sensing techniques,synthetic aperture radar(SAR)applied to the monitoring and assessment of earthquake disasters has unique advantages.SAR allows surface coverage data to be obtained for a large area,is not overly affected by rain or snow conditions,and can quickly obtain earthquake damage information for a large area.Earthquake-damaged buildings can be identified using the information extraction method,although this method has a lack of universality due to different area and atmospheric conditions.A change detection technique using remote sensing images can solve this problem.To extract earthquake-damaged buildings,a change detection technique based on multiple features was developed.The main achievements of this thesis are as follows:(1)The SAR image features of buildings were summarized.Then,the features of the SAR image were analysed,including image statistics features and change detection features.The differences among each feature were compared,and the most suitable features were selected.(2)Earthquake damage information was extracted using SAR correlation and coherence change detection methods.The SAR image of a post-earthquake area was segmented using multi-scale segmentation technology,and based on the segmented object,change detection with respect tocorrelation,coherence and polarization likelihood ratio was performed.The results showed that the accuracy obtained for an object is higher than that fora single pixel.(3)A polarimetric SAR remote sensing damage index(RRSDI)was developed for the case in which full polarization cannot be obtained,and was proposed to calculate the dual polarization SAR damage index.The index has a normal distribution;thus,change detection based on a T-test was performed on dual polarization SAR images.Then,the extraction accuracy of different significance levels was given,and it was found that the extraction effect is stronger when the significance level is set at ?=0.05.Based on the theory that the polarization covariance matrix is a complex Wishart distribution,a full polarization SAR image likelihood model was constructed.The model calculates the likelihood ratio of two full polarization SAR image covariance matrices and then derives a natural logarithm difference image.The difference image was classified using the supervised classification technique.(4)The object feature space based on SAR images was extended,namely using intensity correlation,phase coherence and polarization likelihood ratio information.Using the extended feature space,a method for extracting earthquake-damaged buildings from SAR images based on integrating multiple features and change detection was proposed.Three change detection features of segmented units were integrated using support vector machine classification,and the damaged building information was extracted.(5)Combined with multiple-characteristic SAR image change detection,an extraction pattern for damaged building information using the SAR change detection technique was proposed.