基于核聚类和最优迭代的SAR图像相干斑抑制研究

发布时间：2018-02-06 06:56

本文关键词： SAR 相干斑抑制核回归非局部均值滤波迭代估计　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)作为一种等效天线孔径的雷达,它根据雷达与目标的相对运动来把较小尺寸的真实天线孔径用数据处理的方法进行合成的。SAR图像不仅具有全天候、全天时、分辨率高、可侧视成像等诸多优点,而且包含了丰富的特征信号,其中包括了多种信息,如幅度、相位和极化等,近几年来进行SAR图像的处理(去噪,分割,目标识别等)也越来越受到广泛的关注。但是由于SAR图像的特殊成像原理,会产生相干的散射回波,造成了得到的SAR图像中含有随机的斑点噪声,并且这些噪声是乘性的,使得SAR图像的去噪处理与一般图像不同。这种相干斑噪声影响了SAR图像的质量和后续的处理。因此对SAR图像的相干斑抑制是非常有必要的,并且要尽可能的保留图像的细节信息。SAR图像去噪问题主要就是在去除斑点噪声和保留SAR图像细节这两个方面做到一种好的平衡。本文主要是在更好的保留SAR图像细节信息方向上对相干斑抑制方法做出了一些改进,主要工作和贡献如下:1.提出基于核回归特征聚类和改进非局部均值滤波的SAR图像相干斑抑制方法。主要是通过自适应核回归自身的核函数特性,可以通过权值表示得到SAR图像的一些细节特征(边缘,纹理等)。为了更好的处理这些特征,本文采用聚类的方式,将这些提取的特征作为初始聚类中心,然后利用K-means聚类的方法将相似的特征聚合在一起,这样就可以得到多个具有相似特征的聚类的,接下来通过改进相似性度量的方式进行优化非局部均值滤波,这样有效的保证了在对每一类进行一个非局部均值去噪处理时能够尽可能保留SAR图像的细节部分。2.提出了基于diffusion和boosting的自适应迭代估计的SAR图像相干斑抑制方法。主要是引入了基于最小均方误差(MSE)的一种风险估计,针对diffusion和boosting这两种迭代机制,它们各有优缺点,后者可以弥补前者的缺点,因此本文结合这两种迭代方法的优点,进行自适应选择,得到最优的迭代方法和迭代次数,然后基于这个最优选择得到的结果进行非局部均值滤波,本方法可以在很好的保留图像的细节信息的同时达到去噪的目的。
[Abstract]:Synthetic Aperture radar (SAR) as an equivalent antenna aperture radar. According to the relative motion of radar and target, the synthetic. SAR image with smaller real antenna aperture is not only all-weather, all-day, but also has high resolution. Side view imaging and many other advantages, but also contains a wealth of characteristic signals, including a variety of information, such as amplitude, phase and polarization, SAR image processing in recent years (denoising, segmentation). But because of the special imaging principle of SAR image, coherent scattering echo will be produced, resulting in random speckle noise in the obtained SAR image. And these noises are multiplicative. This kind of speckle noise affects the quality of SAR image and the subsequent processing. So speckle suppression for SAR image is very necessary. The problem of image denoising is to achieve a good balance between removing speckle noise and preserving the details of SAR image. Some improvements have been made to the speckle suppression method in the direction of preserving the details of SAR images. The main work and contributions are as follows: 1. A new method of speckle suppression for SAR images based on kernel regression feature clustering and improved non-local mean filter is proposed, which is mainly based on adaptive kernel regression. We can get some detail features (edge, texture, etc.) of SAR image by weight. In order to deal with these features better, we use clustering method to take these extracted features as the initial clustering center. Then the K-means clustering method is used to aggregate the similar features together, so that we can get multiple clusters with similar features. Then the non-local mean filter is optimized by improving the similarity measure. In this way, the detail part of SAR image can be preserved as much as possible when each class is de-noised with a non-local mean value. (2) based on diffusion and boosting, this paper puts forward a new method based on diffusion and boosting. Adaptive iterative estimation method for speckle suppression in SAR images based on the least mean square error (MMSE) is introduced. A risk estimate for MSE. For diffusion and boosting, they have their own advantages and disadvantages, the latter can make up for the shortcomings of the former, so this paper combines the advantages of these two iterative methods. Adaptive selection is carried out to obtain the optimal iterative method and the number of iterations, and then the non-local mean filtering is carried out based on the results obtained from the optimal selection. The method can preserve the details of the image and achieve the purpose of denoising at the same time.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【参考文献】