Research On Spaceborne InSAR Image Registration And Extraction Of Three-Dimensional Information

Interferometric Synthetic Aperture Radar (InSAR), which is based on SAR and Interferometric technique, is one of the important and potential achievements in microwave remote sensing in recent decades.The characteristics, domestic and international development process and major applications of InSAR are briefly described. Then, based on the theory and data processing of InSAR imaging, the principle and algorithms of image registration and extraction of three-dimensional information, such as flat earth removal, interferogram filtering and phase unwrapping are respectively analyzed and studied deeply with practical examples.Complex image registration is a key step in InSAR data processing, so fast and efficient registration algorithms are a research hotspot in the field. The commonly steps of image registration, several interpolation kernels and quality evaluation methods for interferograms are described, as well as the satellite InSAR image registration algorithms are discussed. Then, a fast registration method based on correlation coefficient is proposed after the analysis of three commonly used registration algorithm, which used frequency-domain correlation coefficient and chirp Z transform respectively for coarse and fine registration, and finally resampled the interpolated slave image by the translation property of FT. Real data were chosen to perform above mentioned algorithms, and meanwhile the runtime and precision were analyzed. The experimental results prove that the algorithm is valid, robust and efficient.The removal of flat-earth phase and filtering are the main steps to improve the accuracy of phase unwrapping, with which the three-dimensional information can be extracted. The flat-earth effect and the removal methods are discussed, and then characteristics of the phase noise in interferograms and the principle of the pivoting mean filter and median filter based on spatial domain as well as Goldstein's filter based on frequency domain are introduced. After that, some discussion is followed with the experimental results of the methods above. The Goldstein's Branch-Cut and the least-square with FFT phase unwrapping methods are compared, with their strengths and deficiencies analyzed. Finally the complete processing of InSAR imaging is shown with real satellite data and relatively coarse digital elevation models are generated.