基于空间信息聚类的SAR图像分割算法研究

发布时间：2018-01-12 12:05

本文关键词：基于空间信息聚类的SAR图像分割算法研究　出处：《合肥工业大学》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)图像分割是SAR自动解译的关键技术之一,其结果对SAR图像的解译具有重要意义。SAR图像特有的相干成像系统提供了较光学图像更高的分辨率,但不可避免的存在严重的相干斑噪声,传统的针对加性噪声的图像分割方法不再适用于SAR图像。Gamma混合模型聚类根据SAR图像特有的统计分布,凭借形式简单、计算方便等优点,已成为当前广泛应用于SAR图像的分割算法。但是该算法仅利用图像的灰度信息,忽略了像素间的空间信息,导致该算法对噪声十分敏感,分割结果往往存在大量的误分现象。为此,本文提出了基于空间信息聚类的SAR图像分割算法,主要工作包含以下两个方面：1)研究分析了一种结合双边滤波与Gamma混合模型的SAR图像分割算法。该算法在模型的参数求解过程中,把每个像素本身所对应的贝叶斯后验概率作为该图像的像素值,通过双边滤波处理,使得图像中相邻像素的贝叶斯后验概率数值得到平滑,获得相近的贝叶斯后验概率数值。在每一次参数迭代过程中,融入一次滤波操作,使得每一次参数求解都受到像素位置的影响,既能准确的分割SAR图像,又能较好的抑制相干斑噪声对分割结果的影响；2)研究分析了采用区域Gamma混合模型的SAR图像分割算法。该算法采用分水岭分割作为初始分割,将过分割得到的区域块作为Gamma混合模型的聚类样本,使得Gamma混合模型聚类由传统的像素级提升到区域级。并进一步考虑区域间的相互关系,引入邻域因子融入到迭代算法中,得到邻域类生成加权概率,提高算法的分割效果。分别针对合成SAR图像与真实SAR图像进行了充分的实验,并进行了较为详细的分析。与传统的Gamma混合模型聚类、区域高斯混合模型以及区域MRF算法相比,本文所提算法的分割结果都分别有了较大的提高,具有更强的抗噪声能力。证明了本文算法在SAR图像分割中的可行性与有效性。
[Abstract]:Synthetic Aperture Radarimage segmentation is one of the key techniques of SAR automatic interpretation. The results are of great significance to the interpretation of SAR images. The special coherent imaging system provides higher resolution than the optical images, but there is inevitable serious speckle noise. The traditional image segmentation method for additive noise is no longer suitable for SAR image. Gamma hybrid model clustering according to the special statistical distribution of SAR image, with the advantages of simple form, convenient calculation and so on. It has been widely used in SAR image segmentation, but the algorithm only uses the gray level information of the image, ignoring the spatial information between pixels, resulting in the algorithm is very sensitive to noise. The segmentation results often have a large number of misclassification phenomena. Therefore, a spatial information clustering based SAR image segmentation algorithm is proposed in this paper. The main work includes the following two aspects: 1) A SAR image segmentation algorithm combining bilateral filtering and Gamma hybrid model is studied and analyzed. The algorithm is used to solve the parameters of the model. The Bayesian posteriori probability corresponding to each pixel itself is taken as the pixel value of the image, and the Bayesian posterior probability value of adjacent pixels in the image is smoothed by bilateral filtering. Obtain similar Bayesian posterior probability value. In each parameter iteration process, a filter operation is incorporated, so that each parameter solution is affected by the pixel position, which can segment the SAR image accurately. It can also suppress the effect of speckle noise on segmentation results. 2) the algorithm of SAR image segmentation using regional Gamma hybrid model is studied and analyzed, which uses watershed segmentation as the initial segmentation. The region block obtained by over-segmentation is used as the clustering sample of the Gamma hybrid model, which makes the Gamma hybrid model cluster from the traditional pixel level to the regional level, and further considers the interrelationship between the regions. The neighborhood factor is introduced into the iterative algorithm to obtain the weighted probability of neighborhood class generation, which improves the segmentation effect of the algorithm. The experiments on synthetic SAR images and real SAR images are carried out. Compared with the traditional Gamma mixture model clustering, regional Gaussian mixture model and regional MRF algorithm, the segmentation results of the proposed algorithm are greatly improved. It is proved that the proposed algorithm is feasible and effective in SAR image segmentation.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【共引文献】