【作者】 李昱彤；

【导师】 周越；

【作者基本信息】 上海交通大学 ， 模式识别与智能系统， 2008， 硕士

【摘要】 合成孔径雷达(SAR)是一种工作在微波波段的相干成像雷达,它在军用和民用方面已得到广泛应用,具有分辨率高,全天候工作,有效识别伪装和穿透掩盖物的优势。然而,其缺点在于雷达接收的散射信号相干性叠加,使成像以后的SAR图存在严重的散斑噪声,极大的影响了图像理解和后续应用。一直以来,成为国内外学者研究的热点。传统方法或采取空域滤波,或基于某特定准则去构造线性滤波器。与之不同,本文着眼于SAR图像多分辨率和信息量大的特点,采用多尺度分析和独立成分分析理论对SAR图像散斑消除进行研究。本文主要研究工作有:绪论部分介绍SAR的研究历史与发展情况,SAR图散斑噪声消除的研究现状,并引出研究内容。第二章系统讲述散斑噪声模型,经典消噪算法,消噪性能评价指标。第三章讲述基于多尺度分析的SAR图降噪算法。主要分为多重分形和小波分析两个部分。包括分形理论基础,点态H?lder指数求取,多重分形谱求取,噪声消除;小波阈值收缩思想,阈值确定,阈值函数设计。第四章讲述基于独立成分分析的SAR图散斑消除算法。包括算法概述,求解模型,经典的稀疏编码阈值收缩算法。第五章讲述基于多尺度分析与独立成分分析相结合的SAR图散斑消除算法。最后一章给出全文总结以及研究展望。全文创新之处在于:第一、从实验学角度,提出使用二值形态学和均值滤波作为辅助技术对实验结果进行优化。第二、提出了基于独立成分分析的自适应空间分离算法。采用信号分离思想,通过阈值分离技术将原图像分为噪声和非噪声两个空间。保留非噪声空间达到去噪目的。采用加权信息熵作为桥梁去拟合“阈值――加权信息熵函数”,从而使阈值得以自动确定。第三、从实验角度,提出基于分形与小波的多尺度内部结合的算法(WF);第四、结合频域处理技术和高阶统计量信息,提出基于小波与独立成分分析结合的算法(WCA);第五、挖掘基图像的信息,提出基于分形H?lder指数的ICA基图像增强分离算法(FCA);第六、结合分形H?lder指数图的良好表征特性,提出基于分形H?lder指数图的编码收缩算法(H-ICA)。更多还原

【Abstract】 Synthetic Aperture Radar (SAR) sensors can produce range imagery of high spatial resolution under all-weather conditions. It can also effectively identify, even penetrate the disguise easily. Therefore, it has been obtained extensive and intensive applications on both military and civil aspects. However, because the SAR image data are formed by coherent interaction of the transmitted microwave with the targets, it suffers from the effects of speckle noise which arises from coherent summation of the signals scattered from scatterers. This kind of noise makes image understanding a very hard job and severely affects its further application. To remove the noise is necessary and has also been obtained great attention from researchers all over the world.Traditional methods are mainly based on spatial filtering or to design linear filters according to specific criterion. Different from which, this paper presents a novel approach using multi-scaled analysis and independent component analysis on SAR image speckle reduction, focusing on both the multi-resolution and great internal information of SAR image.The main research work of this paper concludes:Introduction section presents the current research situation on speckle reduction and the key work of this paper. The second section gives the speckle model of SAR images, typical disspeckling algorithms and assessment indicators on recovery image. In the third section, multi-scaled analysis is presented, in which, multifractal and wavelet are respectively used for SAR image de-noising. Regarding to multifractal theory, some basic knowledge is firstly given, then pointwise H?lder exponent computing and multi-spectral analysis, noise reduction are orderly explained. Regarding to wavelet, the typical thresholding shrinkage algorithm is presented. Its key steps including threshold determination and thresholding function design are also presented. The next section is about speckle reduction based on independent component analysis (ICA), comprised of ICA overview, model solutions and classical ICA sparse coding shrinkage. In the fifth section, presented is de-noising methods taking advantage on multi-scaled analysis and ICA. Conclusions and research prospects are given in the final section.The creative aspects proposed by this paper conclude:Firstly, using binary morphology and average filtering to optimize experimental results.Secondly, proposing adaptive space separation based on ICA. From the view of signal separation, we separate the original image space into noise space and non-noise space by setting proper threshold. Beyond that,“threshold-weighted information entropy function”is interpolated to obtain adaptive threshold.Thirdly, proposing WF (Wavelet--Fractal) algorithm from experimental view, taking advantage of incorporation of both fractal and wavelet.Fourthly, proposing WCA (Wavelet--Independent Component Analysis approach) algorithm taking use of frequency domain technique and high-order statistic information.Fifthly, proposing FCA (Fractal--Independent component analysis) algorithm using basis image enhancement and separation, taking advantage of mining basis image information.Sixthly, proposing H-ICA (H?lder--Independent component analysis) algorithm based on H?lder image coding shrinkage, taking advantage of the good describing ability of H?lder exponent.更多还原