Study On Key Techniques Of Polarimetric Synthetic Aperture Radar Image Applications Based On Visual Cognition

Polarimetric synthetic aperture radar (Pol-SAR) is an advanced radar remotesensing system, which can be used to acquire the Pol-SAR image containingbackscattering and polarimetric information of targets. The ultimate goal of Pol-SARimage application is its interpretation, which can be classified as the humaninterpretation and the machine interpretation. Before the SAR image is interpreted byhuman, it has to be visualized to an image suitable for the observation of the humanvisual system. Two key techniques of the machine interpretation, the terrainclassification and linear feature detection are studied based on the theory of the visualcomputation. This thesis focuses on these key techniques of the Pol-SAR imageryapplication based on the visual cognition, and the innovations are summarized asfollows.1) According to the procedure of the human observation to an image, aframework of the visualization for the single channel SAR image is proposed. Inthis framework, the SAR image of the large-scale region is visualized first, andthen the SAR image of the hot spot regions are visualized based on the datastatistics. The experimental results show that the image of the large scale regionwith high contrast and the images of the hot spot regions with the fine detailscan be obtained.2) The Pol-SAR image can be visualized as a gray image and a color imageas well. Using the visualization method for the single channel SAR image, thevisualization methods for the Pol-SAR image are modified. In addition, newmethods to visualize the Pol-SAR image to a gray image and a color imagerespectively are proposed based on the visual cognition and polarimetricinformation of the data. The experimental results validate the effectiveness ofthe proposed methods.3) Combining the elevation information obtained from the interferometricSAR (InSAR) data, or the height of the trees and buildings retrieved by thePolarimetric InSAR (Pol-InSAR) data and the visualization results of thePol-SAR image, the Pol-InSAR data can be visualized as3-D scene. In addition, this thesis discusses the mechanism of the forest imaging, points out theproblem of the traditional method for the forest height inversion and proposes anew method for forest height inversion.4) The mode of the terrain type classification based on the human system isdiscussed and the procedure of the automatic classification using the machine isproposed based on the theory visual cognition and visual computation. A newmethod of terrain type supervised classification, which is the critical step of theautomatic classification, is proposed for polarimetric Synthetic Aperture Radar(Pol-SAR) images. The proposed method uses an important visual feature, thetexture feature to train the classifer, which is with the binary tree structure. Theexperimental results show that the classification accuracy of the proposedmethod is promoted obviously comparing to the traditional methods.5) According to the visual computation model, the procedure with twolevels, from the coarse level to the fine level, is proposed, which is applicable tothe linear feature detection in both the single channel SAR image and thePol-SAR image. On the coarse level, the coarse regions of the linear features areextracted by a curvelet transform, a multiscale analysis tool with thecharacteristics in accordance with the features of the human visual system. On afine level, a linear feature detector is used to accurately locate the linearfeatures inside the regions. The experimental results demonstrate theeffectiveness of the proposed method.