Study Of Road Extraction For Circular SAR Images

Circular synthetic aperture radar(CSAR)always illuminates the same observation scene during circular motion of the radar platform,which can realize full angle imaging,sub-wavelength resolution,three-dimensional imaging ability and dynamic monitoring of the observation scene,and make it have the ability of fine and comprehensive description of observation scenes.As a result,CSAR has great application potential in the field of road extraction.However,the road in CSAR image appear as curvilinear structue,and traditional road extraction method based on line structure is difficult to obtain good extraction performance.Therefore,this thesis focuses on the key problems of road structure enhancement,edge detection,surface structure road extraction and complex scene road extraction in CSAR image.The main work is summarized as follows:First,the enhancement method of CSAR full aperture image is studied.CSAR full aperture enchanced images are usually obtained by step-by-step registration and uncoherent accumulation of sub-aperture images.However,due to the existence of platform motion error,anisotropic scattering target and terrain fluctuation of observation scene,there are some differences between sub-aperture images,and the differences gradually become larger with the progress of registration and uncoherent accumulation.Therefore,this thesis proposes a registration strategy combining factorized and chain registration,and discusses the applicability of bilateral filter scale invariant feature transform(BFSIFT)registration method based on feature points and the enhanced correlation coefficient(ECC)registration method based on gray value in adjacent sub-aperture images registration.The problem that it is difficult to obtain CSAR full aperture enhanced image due to the weak coherence of adjacent sub-aperture images is solved,which lays a foundation for subsequent road extraction.Second,the edge detection method suitable for CSAR image is proposed.Although CSAR full aperture incoherence accumulation image contains rich,dense detail and texture information,the existing edge detection methods can not obtain complete edge information.Therefore,this thesis first analyzes the ratio of exponential weighted averages(ROEWA)edge detection operator with high edge resolution,and obtains that the image edge presents as bright curvilinear structure in its edge strength map.Then,the curvilinear structure extraction method which can overcome the positioning error caused by asymmetry and strong robustness in a variety of noise enviroments.Finally,the proposed curvilinear structure extraction method is combined with ROEWA to enhance the edge strength,so as to obtain the complete edge information of CSAR.Third,the method of extracting surface strcture road from CSAR image is proposed.The CSAR image provide a detailed and comprehensive description of the scene,showing a variety of features of the road,how to utilize the various features of road is the key to realize the fine road extraction.Firstly,the extraction method of specific polarity curvilinear structure from bipolar image(having both bright and dark curvilinear structure at the same time)is proposed,solving the multiple response problem of curvilinear structure extraction,and the road centerline extraction of CSAR image is completed.Then,based on the centerline and edge information of the road,combining its geometrical and radiometrical features,the geometrical constraint of parallel double edges and radiometrical constraint of dark homogeneous region are proposed to eliminate the most of false alarm.Finally,using the centerline and edge information extracted by the proposed method for extrapolation or interpolation,the missing road edge points and intersection points are extracted,and the extraction of surface strcture road from CSAR image is accomplished.Fourth,the road extraction method for complex scene in CSAR image is proposed.Because of its unique advantages,it is widely used to obtain all-round information of key and hot complex scenes,and a large number of interferences such as buildings and cultivated land are important factors affecting the final road extraction results.In order to surpress a large number of interferences,combining the characteristic that roads appear as dark curvilinear structures in CSAR image,this thesis firstly uses the ranking the orientation responses of path operators(RORPO)to enhance the curvilinear structures.Then,inspired by the push-pull suppression phenomenon of visual neurons,a suppression part is introduced into the bar-shaped combination of shifted filter responses(B-COSFIRE),the curvilinear structures strength extraction method robust suppression-enhanced curvilinear structure operator(RSECSO)is proposed,which enhances the robustness in noisy and large curvature environment.Finally,the interference is further suppressed by hysteresis threshold and morphological processing,and the extracted road information is used for extrapolation to improve the road topology,realizing the road extraction for complex scene in CSAR image.The research of this thesis are verified by real CSAR data,including P,L,Ku and Ka band CSAR images.The experimental results show the effectiveness of the proposed method and the correctness of theoretical analysis.