基于去噪和精英遗传聚类算法的SAR图像变化检测

发布时间：2018-11-23 16:31

【摘要】：图像的变化检测是对同一地区在不同时间获得的两幅图像进行分析,得到变化信息,而由于合成孔径雷达(Synthetic Aperture Radar,SAR)具有较高的分辨率,同时具有全天候和全天时的特点,可方便地获得同一地区不同时间的图像,因此SAR图像已发展成为遥感图像变化检测的主要组成部分。SAR图像的变化检测是对不同时间所得的同一地区表面的遥感图像进行比较得到差异图,然后利用差异图像的灰度值将图像分为变化区域和不变区域。SAR图像的变化检测在环境监测、灾情估计、土地利用、森林采伐监测、农作物生长状况监测等方面有着非常广泛的应用。本文针对现有的SAR图像变化检测方法耗时长和精度低等问题做出了研究,主要内容包括:1.提出了一种基于直方图和精英遗传聚类算法的SAR图像变化检测方法。该方法首先利用直方图的运算减少了图像的样本点数,有效的减少了图像处理的时间,然后利用模糊C均值(FCM)初始化遗传算法的初始种群和计算适应度函数,并运用一种精英策略的选择机制选择出遗传算法的最优解,将该最优解作为FCM的初始聚类中心,结合了FCM的局部寻优能力和遗传算法的全局搜索能力,加快了算法收敛速度,有效的提高了算法的精确度。2.提出了一种基于均值漂移和精英遗传聚类算法的SAR图像变化检测方法。SAR图像处理存在的主要问题是斑点噪声,为了解决这一问题,在该方法中,我们使用均值漂移对SAR图像的差异图进行了去噪处理,有效的降低噪声对变化检测结果的影响。在真实SAR图像的变化检测实验中,经过和一些经典算法的对比,充分证实了该方法对SAR图像变化检测的有效性。3.提出了一种基于改进的非局部均值的SAR图像变化检测方法。该算法主要是针对非局部均值的高斯核加权的欧氏距离对提取图像的特征信息和抑制乘性斑点噪声都不是最有效的,而且使用指数型函数求权重不能很好地处理图像的边缘部分信息的缺点,利用傅里叶核加权的欧氏距离、基于对数比值距离的相似度值测量方法和二维高斯函数的权重,提出了改进的非局部均值算法,并把它用于SAR图像变化检测中,验证了其性能。
[Abstract]:Image change detection is based on the analysis of two images obtained in the same area at different times, and the change information is obtained. Because of the high resolution of synthetic Aperture Radar (Synthetic Aperture Radar,SAR), it has the characteristics of all-weather and all-weather at the same time. Easy access to images of the same area at different times, Therefore, SAR image has developed into the main component of remote sensing image change detection. The change detection of SAR image is to compare the remote sensing images of the same area at different time to get the difference map. Then using the gray value of the difference image, the image is divided into the change region and the invariant area. The change detection of SAR image is used in environmental monitoring, disaster estimation, land use, forest logging monitoring, Crop growth monitoring and other aspects have a very wide range of applications. In this paper, the existing SAR image change detection methods are studied, such as long time consuming and low precision. The main contents are as follows: 1. A change detection method for SAR images based on histogram and elite genetic clustering algorithm is proposed. In this method, histogram operation is used to reduce the number of samples, and the processing time is effectively reduced. Then, the initial population and the fitness function of genetic algorithm are initialized by fuzzy C-means (FCM). The optimal solution of genetic algorithm is selected by using an elite strategy selection mechanism. The optimal solution is regarded as the initial clustering center of FCM, and the local optimization ability of FCM and the global searching ability of genetic algorithm are combined to accelerate the convergence speed of the algorithm. Effectively improves the accuracy of the algorithm. 2. A change detection method for SAR images based on mean shift and elite genetic clustering algorithm is proposed. The main problem of SAR image processing is speckle noise. In order to reduce the influence of noise on the change detection results, we use mean shift to de-noise the difference image of SAR image. In the real SAR image change detection experiment, by comparing with some classical algorithms, the effectiveness of this method for SAR image change detection is fully verified. 3. In this paper, an improved method of SAR image change detection based on nonlocal mean is proposed. This algorithm is mainly aimed at the non-local mean Gao Si kernel weighted Euclidean distance, which is not the most effective for extracting the feature information of the image and suppressing the multiplicative speckle noise. Moreover, using exponential function to calculate the weight can not deal with the edge information of the image well. Using the Euclidean distance weighted by Fourier kernel, the method of measuring similarity value based on logarithmic ratio distance and the weight of two-dimensional Gao Si function. An improved nonlocal mean algorithm is proposed and applied to SAR image change detection to verify its performance.
【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

【相似文献】