Building Detection From High-resolution Polarimetric SAR Images

Target detection and identification using Polarimetric Synthetic Aperture Radar(PolSAR)images are important issues on SAR image interpretation.It is of great theoretical and applicable significance for land cover research,city planning and urban changes monitoring to detect buildings from high-resolution PolSAR images.However,buildings trends to be misclassified as forests and other types because of the complex geometric structure distribution and the cross-polarized scattering of buildings.Therefore,it remains a challenging problem to detect buildings from PolSAR images.This thesis utilized geometric characteristics,texture characteristics and polarization characteristics,which are contained in high-resolution PolSAR images to detect buildings.We also combined texture characteristics and polarization characteristics to detect buildings from the urban areas.The L band ALOS-2 PALSAR-2 fully polarimetric spaceborne PolSAR data and the E-SAR fully polarimetric airborne PolSAR data were applied as experimental data in the thesis.Geometric characteristics,texture characteristics and polarization characteristics were used to detect buildings.Besides,weighted fusion method was also adapted to combine texture characteristics and polarization characteristics to detect buildings.The primary work and achievements of the thesis are as follows.(1)Marker control watershed transform method based on the geometric feature of buildings in high-resolution PolSAR was utilized to detect buildings.Shapes and edges information of buildings showed in PolSAR images are the main geometric feature.Buildings were detected mainly by marker controlled watershed transform method based on the geometric feature of buildings in high-resolution PolSAR images.Experiment results showed that the detection of buildings could keep a good preservation of the edge contour information.(2)Buildings were detected based on the polarimetric scattering characteristics from PolSAR images.In this thesis,the polarization scattering characteristics were consisted of polarimetric target decomposition parameters and circular polarimetirc correlation coefficient.Some common polarimetric target decompositions,such as Cloude decomposition,Freeman decomposition and Yamaguchi decomposition were adapted to extract and analyze the polarimetric target feature in PolSAR images.Besides,considering the fact that circular polarimetirc correlation coefficient is sensitive to buildings detection result,we combined circular polarimetirc correlation coefficient with polarimetirc target parameter to detect buildings through Wishart classifier.Then buildings were recognized from others in PolSAR images.Results indicated that it could detect buildings effectively by taking advantage of polarimetric target decomposition parameters and circular polarimetirc correlation coefficient from PolSAR images.(3)Texture and polarization scattering characteristics were integrated to detect buildings from PolSAR images.Weighted fusion method was applied to fuse texture and polarization scattering characteristics effectively to get the feature set of buildings in PolSAR images.Then a SVM classifier was adapted to classify the surface targets as buildings and others.Thus,buildings were detected effectively from PolSAR images.And the whole detection of buildings were good.In addition,analytical evaluation and comparison of detection results were conducted in the thesis.The experimental results showed that the accurancy and detection rate were improved by combining texture characteristics and polarization scattering characteristics,which was helpful for the detection of buildings from PolSAR images.