双层RBF核快速鲁棒DAISY典型遥感图像配准  

Fast Robust DAISY Based Double RBF Kernel Canonical Correlation Analysis for Remote Sensing Image Registration

作　　者：罗阳倩子[1] 廖威[1] LUO Yangqianzi;LIAO Wei(Guangdong AIB polytechnic college,Guangzhou 510000,China)

机构地区：[1]广东农工商职业技术学院

出　　处：《控制工程》2018年第12期2251-2256,共6页Control Engineering of China

基　　金：广东省教育部产学研结合项目(2012B091000133);广东省工程技术研究中心项目(2012gczxA003)

摘　　要：针对遥感图像配准过程中,存在信息计算量大、精度不高的问题,提出一种快速鲁棒DAISY局部特征RBF核典型相关分析的遥感图像配准算法(FRBFKCCA)。首先。采用快速鲁棒特征提取方式,对遥感图像进行特征描述,并针对该算法存在的精度较低问题,利用DAISY局部特征描述对其进行改进,设计实现了遥感图像的DAISY局部特征描述的快速鲁棒特征提取方法;其次,在遥感图像配准过程中,针对核典型相关分析方法中,存在的逆矩阵计算奇异性难以直接计算的问题,采用双层径向基神经网络实现核典型相关分析的改进;最后,通过实验对所提算法与对比算法进行对比,显示其图像配准准确度可达96%以上,且图像的配准效率和精确更高。In remote sensing image registration process, there existed the information with large amount of calculation, and low accuracy of the algorithm, so, here proposed the fast robust DAISY local feature double RBF kernel canonical correlation analysis of remote sensing image registration to solve this problem. Firstly, here used the fast and robust feature extraction to do the feature description of the remote sensing image, and also used the Daisy local feature description to improved the fast and robust feature extraction algorithm to solve the low accuracy of the algorithm, which design and implementation of the remote sensing image of Daisy local feature description of fast and robust feature extraction methods; Secondly, in the remote sensing image registration process, according to the problem of inverse matrix singularity calculation, here used the radial basis function neural network to achieve the improvement of kernel canonical correlation analysis; Finally, the efficiency and accuracy of the proposed algorithm are compared with that of the contrast algorithm, it shows that the image registration accuracy can reach more than 96 %, and the image registration efficiency and accuracy is higher.

关 键 词：快速鲁棒 径向基神经网络 核典型相关分析 遥感图像 配准 局部特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]