【作者】 纪建；

【导师】 田铮；

【作者基本信息】 西北工业大学 ， 计算机科学与技术， 2007， 博士

【摘要】 合成孔径雷达(Synthetic Aperture Radar，SAR)图像由于其全天侯、全天时、分辨率高等优点，在军事和民用领域中都得到了应用。但是应注意由于SAR成像的机理，使得SAR图像含有斑点噪声，这样与传统光学图像的处理相比，SAR图像的处理具有其特殊性。本文研究基于独立分量分析(Independent Component Analysis，ICA)的方法的SAR图像处理问题。针对SAR图像的统计特性，给出了基于多尺度自回归滑动平均(Multiscale Autoregressive Moving Average，MARMA)模型和基于投影寻踪学习网络(Projection Pursuit Learning Network，PPLN)方法的SAR图像压缩方法；给出了基于独立分量分析的极化SAR图像抑制斑点噪声方法；提出了SAR图像增强的稳健独立分量分析方法；研究了SAR图像滤波和增强的子带独立分量分析处理方法。主要研究内容如下：(1)提出了基于多尺度ARMA模型的SAR图像压缩新方法，以建立多尺度MARMA模型和多尺度MAR(Multiscale Autoregressive)模型来直接刻画SAR图像的统计相依性为基础，据此构造压缩算法。不使用对SAR图像的分割结果，因而该方法无需SAR图像分割的先验知识。【附录二：(1、3)】(2)给出了基于投影寻踪学习网络的SAR图像压缩新方法。该方法综合了投影寻踪回归(Projection Pursuit Regression，PPR)的统计思想和图像压缩的基本思路，试验结果表明该方法既能达到较高的压缩比又能取得较好的保真度。【附录二：(4)】(3)极化SAR图像可视为目标信号与噪声的线性混合，利用ICA的分离性，可从其中分离出期望信号，将HH、HV、VV和HV／VV的比值图像做为ICA的输入数据。本文首次基于ICA方法，得到了分别对应于HH、HV和VV极化的三幅降噪图像，取得了较好的实验结果。进而，还分别使用了三种不同的ICA算法，进行了极化SAR图像的相干斑抑制，并对其结果进行了比较分析，这对补充ICA算法的研究成果是有意义的。【附录二：(5、7、9)】(4)构造了一个稳健独立分量分析神经网络算法。当数据中存在噪声或异常值时，该方法能够有效地降低噪声的影响。该算法通过应用基于最小协方差行列式(Minimum Covariance Determinant，MCD)估计的异常值拒绝法则，使其具有稳健性。在数据中含有较强噪声或异常值时，算法优于传统的ICA算法。将此算法分别应用到盲噪声图像分离和SAR图像处理上，与传统的ICA方法相比，得到了较好的分离结果。【附录二：(2、6)】(5)给出极化SAR图像增强的子带ICA处理方法。子带ICA扩展ICA的基本模型，假设源信号各分量并不具有统计独立性，而是由一些统计独立的子分量组成，与传统算法相比，该方法具有更好的灵活性和鲁棒性结果。本文首次使用基于子带ICA方法进行极化SAR图像增强，得到了较好的实验结果。【附录二：(8、12)】更多还原

【Abstract】 SAR (Synthetic Aperture Radar) images get the rapid development and wide application because they can be obtained any-time, any-weather and with high resolution. Unfortunately, SAR images are always polluted by the multiplicative noise. For this reason, the traditional methods of image processing do not work well.In this paper, a series of new methods are proposed and applied to SAR compression, speckle reduction and enhancement. The main innovative points are as follows:At first, a new method based on multiscale autoregressive moving average (MARMA) models is presented to compress SAR image. The method uses the multiscale representation as the cornerstone of the modeling process, and constructs the MARMA models of image. Thus we predict the initialized image data using these multiscale models, and the compression is subsequently achieved through coding the residual image. Unlike published methods, supervising segmentation for SAR image is not used in our compression processes. So the prior knowledge of segmentation is not required. Experimental results have proven that the proposed method achieves high compression radios with impressive image quality.Secondly, we present a new algorithm for SAR image compression based on projection pursuit learning networks. At first, we segment an SAR image into regions of different sizes based on mean value in each region and then constructing a distinct code for each block by using the projection pursuit neural networks. The process is stopped when the desired error threshold is achieved. The experimental results show that excellent performance can be achieved for typical SAR images with no significant distortion introduced by image compression.Thirdly, the polarimetric synthetic aperture radar (PSAR) images are modeled by a mixture model that results from the product of two independent models, one characterizes the target response and the other characterizes the speckle phenomenon. For the scene interpretation, it is desirable to separate between the target response and the speckle. For this purpose, we proposed a new speckle reduction approach using ICA based on statistical formulation of PSAR image. In addition, we apply three ICA algorithms on real PSAR images and compare their performances. The comparison reveals characteristic differences between the studied neural ICA algorithms, complementing the results obtained earlier.Fourthly, we propose a robust ICA network for separation images contaminated with high-level additive noise or outliers. We reduce the power of additive noise by adding outlier rejection rule in ICA. Extensive computer simulations of separation of noisy image confirm robustness and the excellent performance of the resulting algorithms. In addition, its application in SAR is discussed. The results show the potential usage in SAR image processing problems.Fifthly, the basic model and methods of subband Independent Component Analysis (SICA) system are introduced in this paper. In our model, the assumption of the standard ICA model that the source signals are mutually independent is relaxed. We presume that the source signals are the sum of some independent and/or dependent subcomponents. This blind source separation (BSS) problem is solves by using the subband decomposition as preprocessing of ICA. The method proposed in the paper has been tested for unsupervised separation and enhancement in SAR images. The results indicate that the method is promising for the analysis problem of SAR images. In addition, we use SICA to speckle reduction of PSAR images. The results indicate that the method is promising for the speckle reduction problem of PSAR images.更多还原