Improving the accuracy of image co-registration in InSAR

Abstract

Synthetic aperture radar interferomery (InSAR) has gained attention in many fields because of its accuracy, convenience and efficiency. It has been demonstrated to be a potential tool for generating digital elevation models (DEM). But many critical factors affect the robustness and accuracy of InSAR technology and limit its applications and development. This project is devoted to improving the accuracy of DEM generated by InSAR. Some factors influencing on the accuracy of SAR image co-registration have been systematically investigated. Firstly, a quantitative measure for the quality of interferogram is proposed based on the sum of phase difference (SPD). The simulation results show the variation of the quality of an interferogram can be correctly reflected by the value of SPD. When lacking of well-defined tiepoint for co-registration, the tiepoints are selected in grid form. In order to understand the effects of tiepoint interval on the reliability of co-registration, a series of experiments with different grid sizes have been carried out. The effects of the tiepoint interval and the accuracy of DEM were systematically investigated. The tiepoint interval is not a very sensitive parameter. However, a tiepoint interval with around 200*30 pixels is an appropriate grid size for tiepoint, which can make results more reliable. If feature points are available, more accurate co-registration can be achieved. To extract the feature points, a method is proposed. The extraction of such points is based on the wavelet gradient modulus maximum after decomposing SAR images. The feature points extracted by wavelet and the tiepoints selected in grid form should be combined together for a reliable co-registration. The experimental results show that the reliability of co-registration was improved by the method. Specially, the accuracy of DEM was improved by 16%~35%. The window size is a factor to be considered for tiepoint matching in co-registration. Optimum window size can be determined based on the autocorrelation of SAR image, which indicates the degree of similarity between pixels in an image. With wavelet, the autocorrelation coefficient can be decomposed into different frequency components. By analyzing the variation of amplitude, the optimum window size can be determined. The experimental results show that it is a reliable method to determine the optimum window size by decomposing autocorrelation coefficient with wavelet. Finally, a system for automated SAR image co-registration is implemented. It fuses all techniques developed in this project systematically and efficiently for enhancing the accuracy of co-registration.