利用双差异图和PCA的SAR图像变化检测  

SAR image change detection using double difference images and PCA algorithm

作　　者：刘陆洋 贾振红[1] 杨杰[2] Nikola Kasabov LIU Lu-yang;JIA Zhen-hong;YANG Jie;Nikola Kasabov(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;Institute of Image Processing and Pattern Recognition,Shanghai Jiao Tong University,Shanghai 200240,China;Knowledge Engineering and Discovery Research Institute,Auckland University of Technology,Auckland 1020,New Zealand)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046 [2]上海交通大学图像处理与模式识别研究所,上海200240 [3]新西兰奥克兰理工大学知识工程与发现研究所,新西兰奥克兰1020

出　　处：《计算机工程与设计》2019年第7期2002-2006,共5页Computer Engineering and Design

基　　金：教育部促进与美大地区科研合作与高层次人才培养基金项目(2014-2029、2016-2196)

摘　　要：为减少噪声对SAR图像变化检测分类结果的影响,提高最终的检测精度,提出一种结合双差异图与PCA-FCM的SAR图像的变化检测算法。通过对数比法和均值比法得到同一地区的两幅差异图,两幅差异图通过合成规则得到最终的差异图,将合成的差异图分成不重叠的块,利用主成分分析(PCA)提取特征向量,将合成差异图中每个像素用映射到特征向量空间的向量表示,使用模糊C均值聚类对每个像素的向量聚类分为变化类和未变化类。实验结果表明,该方法降低了噪声对检测结果的影响,提高了变化检测的精度。To reduce the impact of noise on classification results of SAR image change detection,a SAR images change detection algorithm based on double difference maps and PCA-FCM was proposed. Two difference maps in the same region were obtained using the log-ratio method and the mean-ratio method. The final difference image was obtained using the log-ratio difference image and mean-ratio image through a synthesis rule. The final difference image was divided into non-overlapping blocks,and principal component analysis (PCA) was used to extract eigenvector of each block. Each pixel in the final difference map was represented by a vector that was projected into the eigenvector space. The vector of each pixel was classified into change pixel and unchanged pixel using fuzzy C-means clustering. Experimental results show that the proposed method reduces the effects of noise and improves the accuracy of the change detection.

关 键 词：差异图 SAR图像 变化检测 主成分分析(PCA) 模糊C均值聚类(FCM) 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]