Research On Built-up Area Information Extraction From SAR/PolSAR Imagery

Along with the rapid urbanization on the world,urban remote sensing techniques have attracted more and more attention in recent decades.Synthetic Aperture Radar(SAR)and Polarimetric Synthetic Aperture Radar(PolSAR)have been widely used in remote sensing due to their imaging capabilities of providing robust and rich information in almost all-weather and solar illumination conditions.This thesis focuses on the built-up area information extraction from SAR and PolSAR data,which consists of scattering mechanism analysis,built-up area detection,segmentation,and classification.The objective is to provide a whole chain for urban area information extraction from radar remote sensing data,which can be used as the technical support for urban planning,economy development and military applications.The main work of this thesis is as following:(1)Backscattering and polarimetric scattering analysis of buildings,which will be used to develop polarimetric target decomposition algorithms.The backscattering characteristics of buildings in SAR imagery and polarimetric target decomposition theory are firstly reviewed.Then the challenge of polarimetric decomposition over urban areas,that is scattering ambiguity between buildings not parallel to the radar azimuth direction and vegetation,is analyzed.On the basis of current decomposition techniques,this thesis proposes a cross scattering coherency model,which considers the building orientation and scattering mechanism.This scattering model can effectively describe the HV scattering of buildings and make it different from that of forests,thus the scattering ambiguity can be avoided.After that,multiple-component and four-component polarimetric decomposition techniques are proposed on the basis of this cross scattering model.Spaceborne and airborne PolSAR data are used to demonstrate the effectiveness of this scattering model.Finally,the differences between these two decompositions are compared and their potential abilities on building detection and classification are also discussed.(2)Built-up area detection from SAR and PolSAR imagery.Firstly,we give the brief overview of textural-based building detection method for SAR imagery and the azimuth nonstationarity detection method for PolSAR imagery,as well as the limitations of these two methods.To remove the false alarms of forests and further improve the building detection accuracy,this thesis proposes a new building detection method for PolSAR imagery,which considers the azimuth nonstationarity and reflection asymmetry of man-made targets.These two scattering characteristics of buildings are taken into consideration during the detection,leading to a significant removal of the natural areas.Spaceborne and airborne PolSAR data are used to validate the effectiveness of this method.In addition,on the sub-aperture decomposition and multiple-component decomposition,this thesis further proposes another built-up area detection method,which uses the double-bounce scattering power,cross scattering power and average polarimetric coherence ratio.The experimental results with spaceborne PolSAR data demonstrate that this method not only can remove the false alarms of forests and mountains but also can further improve the detection accuracy of buildings with different polarization orientation angles.These building detection methods for SAR and PolSAR data are finally compared.(3)Built-up area segmentation in SAR and PolSAR imagery.Segmentation of built-up areas can provide useful information for building detection and classification and also can be used in speckle suppression and estimation of urban areas.This thesis firstly reviews the state-of-the-art building segmentation methods for SAR and PolSAR imagery and then proposes a new fuzzy segmentation approach for SAR imagery.To obtain the building segmentation details from high resolution SAR imagery,a superpixel generation method is proposed and applied in urban area segmentation.For high resolution PolSAR data,we firstly propose an edge detection algorithm,which can effectively extract the building edges of complex urban areas in PolSAR data.With these edges,a new pixel similarity measure is defined and an effective superpixel generation method for PolSAR imagery is proposed.Segmentation results of high resolution airborne PolSAR imagery validate the effectiveness of this proposed superpixel generation method.From the segmentation results of SAR and PolSAR data,it can be concluded that superpixel can acquire detailed structure information of buildings,which will be beneficial for built-up area classification.(4)Built-up area classification in SAR and PolSAR imagery.Traditional K-means unsupervised classification method based on Wishart distance for PolSAR imagery is firstly introduced,as well as its deficiency on the classification of urban areas.To discriminate the buildings with different polarization orientation angles,on the basis of previous multiple-component decomposition technique,this thesis proposes a scattering power-based K-means unsupervised classification method for PolSAR imagery.Spaceborne and airborne PolSAR data are used to demonstrate the effectiveness of this method.In addition,we also implement the superpixel-based classification for SAR and PolSAR imagery with the help of superpixels.Comparison between pixel-based and superpixel-based classification results indicate that the proposed superpixel can be regarded as the unit in SAR/PolSAR object-based processing,which can improve the urban area classification accuracy of radar remote sensing imagery.