Research On SAR Imagery Terrain Correction And Application

Synthetic Aperture Radar (SAR) can acquire the scattering signals of land cover under different conditions through the coherence measurement, which can provide rich scattering and polarimetric interferometric information for natural and man-made targets interpretation. However, the complicated scattering mechanism and terrain undulation effects have posed challenges for SAR imagery interpretation since the geometric distortion caused by the topographic, which has led to the foreshortening, layover and shadow in SAR image. In addition, the accurate backscattering signal retrieval should depend on accurate terrain parameter such as the local incident angle which is significant for the SAR scattering mechanism, where radiometric distortion may hinder the precise SAR image interpretation. The purpose of this study is to develop the terrain correction method for improving the SAR images interpretation by analyzing radiometric and geometric distortions induced by the terrain. Then, based on the terrain correction method, the SAR interpretation methods for the natural and man-made target interpretation are investigated. The content of the dissertation are as follows:(1) The principles of the terrain induced geometric and radiometric distortions during SAR imaging are summarized and a detailed review on the state-of-the-art geometric and radiometric methods has been presented. The geometric correction algorithms have included the GCP based polynomial, collinearity equation and the DEM simulation based RD approaches. This dissertation has also reviewed the radiometric correction algorithms including the scattering area based radiometric normalization method and the incident angle based cosine function method. Then, the study has presented the real SAR datasets experiments using these different methods to characterize the advantages and disadvantages, which is important for the further algorithm improvement.(2) Current geometric correction algorithms have demonstrated the efficacy in the geocoding for SAR images, but the inaccurate radar backscatter value brought by the terrain has not been paid enough attention to its investigation. The incident angle based radiometric correction algorithm has ignored the relationships between the topologies of radar slant range and ground range, which is the notorious nature of one to many and many to one. The bright lines are brought by local radiometric distortion in the scattering area based method due to different scattering mechanisms of different land cover. As a result, a novel scattering classification based method is proposed for solving these problems by improving the estimation of backscattering coefficients via characterizing different scattering mechanisms using the phase differences in the covariance matrix. Meanwhile, ground control points (GCPs) are used to improve the fitting accuracy of scattering center in addition to the look up table (LUT) based matching approach. We also modeled the relationship between the incident angle and backscattering to improve the accuracy of backscattering coefficient in pixel cells. In the experiment section, mean value, standard deviation, slope between the incident angle and backscattering coefficient, and the correction ratio are used to evaluate the performance of different methods. In combination with the different resolution of DEM to the terrain correction result, the crop's variance of backscattering coefficient during the growth has been analyzed.(3) In natural land cover interpretation, forest parameters inversion have been investigated based on the improved dataset using the terrain correction method. With respect to the forest parameter extraction procedure, PolInSAR based forest classification, forest height inversion as well as biomass estimation is the key problem in PolInSAR forest remote sensing. The dissertation first introduced the volume scattering and surface scattering model. In the two-layer scattering model including the RVoG and OVoG, the phase difference of the corresponding channel (ESPRIT), and the model based height inversion method including the non-linear iteration, three-stage inversion have been investigated. However, the existing algorithms have not adequately considered the different characteristics of the different vegetation distribution and scattering, which have limited the improvement in the accuracy of the forest inversion. In response to the limitation, this dissertation proposed a diversity estimation method. First, the radiometric based terrain correction procedure is used to eliminate the effect of the topographic on forest vertical parameter inversion. Then, the height inversion models for coniferous and broad-leaved forest are constructed separately for improving the accuracy of height inversion, which is significant for the biomass, Leaf Area Index (LAI) and forest type estimation, and the reference information of Primary Productivity (PP) in forest research area.(4) In artificial target interpretation, with the improvement of image resolution, SAR image based target interpretation has been a challenge since the feature consists of simple categories in the original SAR image may turn into inconsistence among the complex categories. Especially under the intense speckle noise, the high interclass and intra-class backscatter variability degrade the image details and decreases interpretation ability. In this work, we focus on the target detection and location using the proposed terrain correction method. First, we analyze the statistical distribution based CFAR approach and Part-based Model target detection method. For reducing the speckle noise of SAR images, a Compressing Sensing (CS) based method was also proposed, which takes the advantage of orthogonality of different observation vectors to suppress the speckle, so as to improve the accuracy of target detection. Since the structure characteristics can provide rich information for high resolution scenes, a statistical distribution based Part-based Model algorithm was proposed. This method makes use of a combination of multi-components structure model and statistical features in order to improve the accuracy of target detection. Finally, in urban area, buildings under high resolution and high incident angle condition may cause upside down phenomenon. In the dissertation, based on the proposed terrain correction procedure, a geocoding based multi-component model is designed to recognize the location of the top and bottom of the buildings, which can provide important auxiliary information for target recognition and image modeling.