Interferometric Synthetic Aperture Radar (InSAR) has been developed rapidly and used widely in recent years. Subjected to time, terrain and baseline factors, the obtained DEM accuracy is not high enough. When the baseline is short, the interferometric fringes is relatively clear, but is not sensitive to changes in elevation, resulting in low accuracy of obtained height value; when the baseline is longer, more sensitive to the changes in elevation, in theory higher elevation accuracy can be achieved, but the situation for a longer baseline, due to overlap, noise, loss of coherence and other factors, effectively interferogram filtering and phase unwrapping are difficult. Combining the advantages of short baseline and long baseline, the Multi-Baseline InSAR can improve the precision of altitude measuring. Maximum likelihood estimation (MLE) is a classical phase estimation method. The difficulties of Airborne Multi-Baseline InSAR are images registration and baseline estimation.
First, for the transformation between two or more images caused by the instability of flight platform, this paper presents a partial registration method which combine SIFT method and an analytic searching sub-pixel registration method. Each flight cannot be exactly at the same velocity, direction and attitude, so pan, zoom, rotate, or even irregular local deformation are generated between every couple of images. Having strong matching ability and good robustness, SIFT method can accurately select the feature points, thus completing the coarse registration. Based on coarse registration which achieves pixel-level precision, the images are block processed, by using of analytical cost function, searching the sub-pixel offsets in the continuous domain. As a result the registration accuracy is improved, and it achieves a perfect accuracy in interpolating sense.
Next, high precision POS, interference pattern itself and the registration parameters are used to calculate the baseline parameters.
Then, the interferometric phase is estimated by using the MLE method. MLE is a classical method to estimate the interferometric phase. Assuming that ground zero-mean white Gaussian random process, construct the likelihood function. Use airborne SAR SLC images and MLE algorithm to calculate the optimal estimation by numerical analysis methods. Both the phase information and amplitude information of SAR image are used, increasing the non-ambiguity range of the interferometric phase, reducing the scope of InSAR blind corner, to improve the accuracy of the interferometric phase unwrapping.
This paper select SAR image covered fields in Henan Province of China obtained by CASMSAR. CASMSAR is an airborne SAR system with an X-band dual-antenna and a P-band antenna developed by a team led by Chinese Academy of Surveying and Mapping (CASM). In June 23, 2011, CASMSAR flew over Dengfeng region three times, with distances between routes being separately 25 meters and 250 meters. Experimental area is hilly areas, undulating terrain. P-band SLC data were processed to get interferogram and baseline parameters by using the above methods. Through removing the flat-earth effects, filtering interferogram, phase unwrapping in traditional methods, at last, the DEM was generated. This experiment shows that registration accuracy and coherence are improved, the baseline parameters are calculated more accurate, compared with single baseline InSAR DEM generated by multi-baseline InSAR can achieve higher accuracy.
In this paper, SIFT method and an analytic searching sub-pixel registration method are combined to solve registration problem of large deformation between images, and accurately estimate interferometric phase by using MLE method. At last, a more accurate DEM was generated.