Synthetic Aperture Radar Interferometry (of Insar) Three-dimensional Imaging Technology Research

Synthetic aperture radar interferometry (INSAR) imaging technique is a new space observation technique developed in the recent years. It uses the phase difference of the radar returns in two complex synthetic aperture radar (SAR) images, of the same area acquired at separate viewpoints or times, to obtain the third dimension information of targets in the surface. In this dissertation, we first review the development of INSAR imaging technique. Then, according to the features and the data processing procedure of INSAR, the key technique including complex images coregistration, interferogram denoising and phase unwrapping are studied in detail. The relative digital elevation models (DEM) of imaging areas are generated and the main factors that influence the mapping accuracy are discussed.In this dissertation, neural network technique is introduced into complex images coregistration, and a new method of INSAR complex images coregistration using correlation matching combined with neural network is presented. Firstly, the correlation matching algorithm is used to complex images coarse coregistration. Then, neural network is utilized to complex images fine coregistration to produce subpixel accuracy. Experimental results show that the proposed method can be used in INSAR complex images coregistration.By analyzing the noise sources and the characteristics of interferogram, interferogram denoising methods based on fuzzy neural network are proposed. In the methods, interferogram denoising is divided into two steps: noise detection and noise filtering. Firstly, B-spline function is used as fuzzy membership function and a noise classifier using fuzzy B-spline basis function neural network is introduced. Secondly, wavelet basis function is used as fuzzy membership function and a noise classifier based on fuzzy wavelet basis function neural network is presented. In the classifier, the shape of membership function can be adjusted in real time. It endues the classifier with better capability of learning and self-adapting. Thirdly, by researching the principle of cerebellar model articulation controller (CMAC) neural network and its drawbacks, fuzzy theory is imported into CMAC and a noise classifier using fuzzy CMAC neural network is developed. It can better reflect the fuzziness and continuityof human cerebella. Finally, a select multi-mold adaptation median noise filtering method is proposed. Experimental results show that the three proposed interferogram denoising methods can reduce interferogram noise efficiently as well as preserve edge and detail very well.Based on the principle of INSAR phase unwrapping, branch cut algorithm and least square algorithm are researched. Then, a minimum spanning tree phase unwrapping method is presented. Idea and procedure of the method are described in detail. Experimental results show that the proposed method can keep the continuity of phase unwrapping results better.Using the INSAR data processing software developed by ourselves, experiments have been carried out based on ERS-1/2 SAR tandem data and the results demonstrate the feasibility and validity of the methods proposed in this dissertation. The main factors that influence the INSAR mapping accuracy are also discussed in this dissertation.