【作者】 池明旻；

【导师】 黄盛璋；

【作者基本信息】 厦门大学 ， 无线电物理， 2002， 硕士

【摘要】 随着航天航空技术的飞速发展，合成孔径雷达(SAR)不仅广泛应用在军事上，而且在农业、地理、海洋、气象等领域也有广泛的应用。但是由于SAR是干涉成像，实测得到的图像几乎被speckle噪声完全淹没。如何从严重被污染的图像中恢复出有用信号——即SAR图像去噪增强处理，一直以来都是遥感图像处理的一大难点，同时也是一个热点[21]。 小波的多分辨分析(MPA)特性，在降噪过程中使它既可以有效地抑制噪声，又可以很好地保持图像的大体轮廓特征。Mallat,Witkon,Donoho[12，13，14]等几位小波学家也先后提出了小波域内去噪的技术，尤其是1995年Donoho提出小波软阈值去噪方法，由于它的简单有效，一经提出就得到了极大的推广。但SAR图像信息在各频带上分布的复杂性，及噪声模型的不可确定性，如果直接使用Donoho理论在二维的推广对speckle噪声进行平滑，则不仅去噪不彻底，而且在降噪过程还会严重损失高频有用信息。 本文将小波软阈值方法引入到SAR图像去噪处理中来，并对它进行了改进，提出自己的阈值公式。所做的工作如下： 1．将Donoho的小波软阈值去噪方法(本文称之为小波全阈值方法，WGST)推广到二维图像信号处理，对SAR图像进行了详细地分析并给出去噪和相应特征提取处理后的实验结果。 2．由于噪声能量在不同方向(水平、垂直和对角)的高频系数分布情况有所差异，对小波全阈值方法进行了改进之处是对不同尺度高频系数取不同阈值，同时对同一尺度不同方向的高频系数也取不同阈值，本文称之为小波局部软阈值(WLST)。 3．在文中对speckle噪声做Gaussian近似，提出了改进方法所用到的阈值公式。并把乘性噪声通过对数变换转为加性的，再使用改进后的算法对其去噪。 实验结果和数据分析表明，改进后的小波软阈值方法对SAR图像的去噪增强和大体轮廓保持很有效，因此可将它应用在带有speckle噪声的SAR图像去噪处理。更多还原

【Abstract】 With the raPid development of sateilite astronomy technique, Syntheticaperture radar (SAR) is wde1y app1ied not only in military field, but aiso in theagricuitural, geograPhic, oceanic, weather etc. fields. However, for itsinterterence imaging, observed images are contaminated by specMe noise. It isa1ways a hot yct hard topic for effectively suppressing speck1e noise of SARimage in the remote sensing image processing.The multi-resolution analysis (MRA property of the wave1et can make it notonly effectively reduce noise, but a1so we1l suited preserve edge structure.Waveletist such as Mallat, Witkln and Donoho propo$e the wavelet de-noisingmethods. Especia11y, the wave1et g1obal soft-thresho1ding (WGST) de-noisingapproach put formrd by Donoho in 1995, for it is simple yet effective tode-noise the additive white gaussian noise, has been wide1y spread. However,the signal of SAR image is complicatedly di$tributed in every subband, and themodel of sPeckle noise is uncertain. So direct1y using 2DItwo--dimension)extending a1gorithm of the 1D(one dimension) soft-thresho1ding for smoothingspeck1e noise, noise cannot be effective1y reduced, and also, high-frequencystructure wtll severely be lost.In the paper, our research work introduces WGST algorithm to SAR imagede-noising for speckle noise, and improve it to be aPplied in the remote sensingimage processing (improved WGST, ca1led wave1et 1ocal ST, WLST). At the sametime, propose a thresho1d formula for the WLST algorithm. Our research work as2xl7X#,%II8g@tf&&zfollows:Firstly, We otend the WGST algOrithm from 1D signa1 to 2D one, andcarefully analyze the SAR images, and then give the de-noised andedge-edracted eXPerimental resu1ts using the 2D WGST.Second1y, because speck1e noise is distributed complicatedly in themulti-resolution and mu1ti-orientation high-frequency coefficients, an improved, algorithm is put forward, which lies in selecting 1eve1- o ri e nt ati o n- d ep en d entthresholds.Finally, Assuming speck1e noise as a near gaussian model, a novel thresho1dformula, well suited to the improved algorithm, is proposed. And themultiplicative speckle noise is transformed into an additive one by thelog8rithmic transformation periormed on SAR image, and then the WLSTa1gorithm is applied for speckle noise reduction.The experimental results and data ana1ysis show that it is effective for WLSTa1gorithm app1ied in SAR image for speck1e noise removal and edge preserving.Hence the improved algorithm WLST can be effectively used in SAR image withspeckle noise.更多还原