摘要 近年來，利用合成孔徑雷達干涉相位來獲取地表高程資訊已經是一項很成熟的技術，要獲取干涉相位的方法有兩種，一種是dual-pass，利用雷達在不同時間通過同一地區，例如：衛星; 另一種是dual-antenna，指一個平台上有兩個雷達，同樣也可獲取干涉相位，例如：AIRBORNE。干涉相位的來源主要有地球曲率、載具姿態、地形效應、電離層延遲、對流層延遲，另外還有干涉相位的雜訊。由於此雜訊會增加相位解套困難度，目前在於雜訊的濾除研究已經發表了很多方法，大部分都是利用視窗(window)處理或是平均(average)處理來達到減少雜訊目的，但是其空間解析度都會下降。最近，小波轉換的技術受到各個領域的重視，透過二維的小波轉換可將影像分成低頻部分及不同方向的高頻部分，藉此特性可以把訊號與雜訊分離達到濾波的目的。在本研究中，首先針對此干涉相位雜訊加以討論其特性及統計意義，並引入小波轉換及空間相關器(spatial correlator)來減少雜訊及保留較好的空間解析度。 Abstract Synthetic aperture radar interferometry (InSAR) has become a useful technique to obtain information about the slant range structure of terrain. In the last years, there are many research in this topic. Most of all papers attempt to solved the interferometric phase noise in spatial domain and some in frequency domain. The interferometric technique is based on taking two SAR images in complex. It can be taken from two slightly different positions of the same area, and the generating an interferogram. There are two ways to get the interferometric phase. The first way is single-pass interferometry, where the images are taken at the same time by two antennas separated by a baseline in the cross-talk direction (two-antennas). The second way is repeat-pass interferometry, where the platform carrying the sensor travels over the same area two times with slightly different paths (two-paths). The interferometric phase is due to the interaction between two SAR images. The accuracy of the interferometric phase depends on different factors, but the most important is the coherence. The spatial decorrelation and the incident angle are parameters that reduce the coherence between the images. For repeat-pass interferometry, the temporal decorrelation is a source of coherence loss. Another parameter that affects the DEM quality is the baseline or separation between the antennas. The higher the baseline the noisiest the interferometric phase, as the speckle pattern is ore decorrelated between the images. The statistics of interferometric phase have been characterized by a probability density function (PDF) based on the circular Gaussian assumption. The interferometric phase PDF depends on the coherence and the number of looks. In order to denoise, the improvement in interferometric phase is based on improving the coherence between both images. And the standard deviation of interferometric phase PDF is also significant. There exists several filters that remove the interferometric phase noise. The simplest one is the box car filter that makes a simple multilook or averaging. Another used filter is the two dimension Gaussian filter. These filters do not adjust to the noise level variability. All these filters have a common point : windowing processing (or family of windows). There has been growing interest in despeckling SAR images using wavelet multiscale techniques recently. The speckle effect in SAR images is characterized as multiplicative random noise, whereas most of existing wavelet denoising algorithms were developed for additived white Gaussian noise (AWGN), as AWGN is common in imaging and sensing systems. In this study, we use the AWGN for the speckle noise to simulate interferometric phase. When the problem of interferometric phase reduction is addressed, the following points have to be take into account: to maintain the spatial resolution of original image, avoid the phase jumps in order to make possible the unwrapping process, and keep the fringe pattern of interferometric phase and target information. For the property of wavelet decomposition, we presented the wavelet transform to denoise the in interferometric phase in the complex domain. And also takes the spatial correlator into our algorism to preserve the target and edge information.
