BUILDING AREA EXTRACTION FROM POLARIMETRIC SAR DATA VIA STATIONARITY DETECTION AND CIRCULAR-POL CORRELATION COEFFICIENT

摘要

Since the buildings have complex geometries and mayrnbe misclassified as forests or mountains with volumernscattering due to the significant cross-pol backscatterrnand lack reflection symmetry, especially the slantorientedrnbuildings, building area extraction is arnchallenging problem. In this paper, the time-frequencyrndecomposition technique is adopted to acquirernsubaperture images, which correspond to the same scenernresponses under different azimuthal look angles.rnStationarity detection approach with polarimetricrn0 G distribution is proposed to extract ortho-orientedrnbuildings and the circular polarization correlationrncoefficient is optimal in characterizing slant-orientedrnbuildings. We test the aforementioned method usingrnESAR image with L-band. The results demonstrate thatrnthe proposed method can effectively extract both orthoorientedrnand slant-oriented buildings and the overallrndetection accuracy as well as kappa value is 10%-20%rnhigher than the compared methods.