Geocoding and stereoscopy of synthetic aperture radar images

This thesis is concerned with the geocoding of Synthetic Aperture Radar (SAR.) images and the use of stereo SAR images. The work was carried out as part of the preparation for the launch of the ERS-1 satellite, due in July 1991, which will carry a SAR sensor. There are two basic approaches to geocoding: image-to-object and object-to-image. Both of these methods have been analysed and assessed on experimental data, namely SIR-B imagery of Mount Shasta. Each type of geocoding requires the solution of nonlinear equations. It has been shown that if the parameters which control the geocoding process are given to a good degree of accuracy, each method can give good results. The effect of inaccuracies in the estimation of these parameters has also been analysed. It was found that there was a predominantly linear response to parameter error in both types of geocoding. Experimental investigations into the effects of the resampling, inherent in operational geocoding, showed that the statistical properties of the resulting image may be severely corrupted with pixel values of less than zero being obtained. This discovery has subsequently been given theoretical support. Height can be determined from stereo pairs of images and digital elevation models can thus be produced, aiding both geocoding and topographic mapping. Existing approaches to SAR/SAR stereo all appear to be based on photogrammetric methods. An alternative, analytic approach, believed to be novel, is described and applied to the same Mount Shasta imagery. Using this method, with accurately-known controlling parameters, correspondence with ground data is excellent. However, an analysis of the sensitivity of the approach to inaccuracies in the controlling parameters shows that the method is extremely sensitive to error. The possibility of combining SAR and optical/infrared imagery for stereometric purposes is also discussed from a theoretical viewpoint