Noise Filtering For InSAR Interferometric Phase Images

Radar has the advantages of active, all weather and 24 hours' operation to the observed target. Synthetic Aperture Radar (SAR) is a new Radar system technique which can acquire signal from the earth surface and form images with very high resolution. It has been widely used on military and commercial aspects. Compared with visible and inferred sensors, SAR can not be disturbed by weather and also the signal can penetrate clouds and forests。 In addition, longer wavelength has the ability of penetrating some particular materials underground and therefore can be used to do underground detection/observation. Interferometric SAR (InSAR) is one of the most important developments of SAR technique. It can be used to generate Digital Elevation Model (DEM) so that three-dimensional surveying of earth's surface can be achieved. From 70's in last century, InSAR technique has been being developed rapidly not only in theory but also in practice. In terms of its applications, there are lots of significant achievements in topographic surveying, earthquake monitoring, volcano monitoring, deformation detection, etc. Well-developed high-resolution SAR and InSAR has been produced and used by U.S., Canada, Germany, etc. In China, Radar imaging has also been put into key project in some plans, such as "863" high-tech study and research plan.In InSAR process, the most important step in the chain process is phase unwrapping, Its quality is controlled directly by the quality of the interferogram. If the quality of interferogram is poor, the result of phase unwrapping will not be satisfactory, sometimes even impossible to do phase unwrapping. Therefore, the interferogram must be pre-processed to make sure the accuracy of DEM. Because of the spatial decorrelation, which is the baseline between the two satellites, and the temporal decorrelation, which is the time difference between the two acquisitions, lots of noise occurs in the interferometric phase images. Other noise, such as thermal noise and speckle noise, also exists in the interferogram. The interferometric phase images must be filtered before phase unwrapping to improve the accuracy of DEM. In some of the papers, coherence has been used as a veryimportant parameter for the interferogram filtering. Coherence map is actually a byproduct generated from the two SLC SAR images. Each pixel in coherence map has a corresponding pixel value in the interferogram which is generated based on the same two images. In addition, coherence has been proved to be highly correlated to the noise level of the interferogram, i.e., the phase with maximum coherence value 1 has no noise in the pixel, whereas the phase with the minimum coherence value 0 is totally corrupted by noise. Therefore, it is used to assist in the new proposed interferogram filtering techniques to remove the noise adaptively.There are several techniques having been proposed to remove phase noise in the interferogram. A widespread tool is the mean (or boxcar) filter that applied in the complex plane and makes an averaging effect on a moving rectangular window of, e.g., 7x7 pixels. But its simplicity also results in over smoothing images so that more details of the fringes will be destroyed and precision of the DEM will be degraded. Another popular filter, median filter, calculates the median value of a neighborhood on real part and imaginary part of the complex separately and as a consequence gets more robust average value than the boxcar filter such that a single very unrepresentative pixel in a neighborhood will not affect the median value significantly. However, it does not adapt to local noise level variations. In Lee's technique, 16 directionally dependent windows have been applied to preserve the fringe pattern/phase gradient. But the improvement has been shown at the expense of computational complexity. An example of filtering algorithms applied on interferometric phase images in frequency domain is Goldestein filter. All these techniques however involve loss of image detail to a certain extent.In order to degrade the noise in the interferogram while preserving the detail information in it as much as possible, a new filtering technique has been proposed. It consists of two stages of filtering technique: The first one is mainly separating the signal into smooth part (low frequency part) and non-smooth part (high frequency part);the second one is filtering out the noise component in the non-smooth part and preserving the detail to a certain extent. At last, it is add back to the smooth part to get the final output.In the thesis, several typical interferogram filters and the new proposed filter are described and compaired, both simulated and real data has been used to evaluate the performance of the filtering techniques. The results show that the new proposed technique not only reduce the number of residues the ease the phase unwrapping step, but it also provide a higher quality DEM profiles among all the typical filtering techniques mentioned when compared against ground truth data.