基于NSCT域图像融合改进算法的SAR图像无监督变化检测

Unsupervised detection of different SAR images based on improved NSCT domain ima ge fusion algorithm

摘要:

为了降低合成孔径雷达(SAR)遥感图像变化检测总 误差数以及获取更好的变化检测结果,提出了一种基于非下采样轮廓变换(NSCT)域图像融合 改进算法的SAR图像无监督变化检测方法。首先用两时相遥感图像构造得到对数比值 和均值比值两种差异图,并 经过NSCT分解得到各自的高低频系数；然后对两差异图中的低频系数均采用最大梯度平方和 规则进行融合,高频 系数均采用加权规则进行融合,再利用逆NSCT融合得到最终差异图；最后用模糊局部信息C 均值聚类(FLICMC)算法实现对图像变化区 域与非变化区域的分类。通过对真实遥感数据集与模拟遥感数据集的研究表明,本文算法能 够有效抑制斑点噪声对 检测结果的影响,并且不受变化类和非变化类统计分布的限制,不需要先验知识,适用性强 ,可以得到较好的检测 效果。与DWT2-FLICMC、MRF-FCMC算法相比,本文算法具有更高的检测精度和较少的算法 运行时间。

Abstract:

In order to reduce the total errors and obtain a better result in change detection with different SAR images, an unsupervised change detection algorithm is proposed in this paper.The techn ique is based on nonsubsampled contourlet transform (NSCT) domain with an improved fusion algorithm and fuzzy local inform ation C-means clustering (FLICMC). First of all,mean-ratio operation and log-ratio operation are adopted to obta in the difference image,by using two images acquired in the same geographical area at different moments.Secondly,the two di fference images are transformed into NSCT domain.The low frequency coefficient s of two images of difference images respectively are fused by the rule of the maximum gradient quadratic sum; the hi gh frequency coefficients of two images of difference images respectively are fused by the rule of the weighted fusion.And the final difference image can be got by using NSCT inverse transform.Finally,on the basis of FLICMC algorithm,the fin al difference image is divided into change area and unchanged area.According to the research between the real remot e sensing data set and the simulated remote sensing data set,the experimental results show that the proposed algorit hm can effectively suppress speckle noise impact on the detection results,and is not limited by the statistical distribut ion of change class and unchanged class.It has great applicability and doesn′t require prior knowledge.So the proposed algorit hm can get better detection effect.Compared with DWT2-FLICMC and MRF-FCMC,the proposed algorithm can obtain higher detection accuracy and shorter running time .