Current Issue Cover
SAR图像的快速核主成分分析识别方法

于红芸,姜涛,关键(海军航空工程学院信息融合研究所, 烟台 264001;鲁东大学数学与信息学院, 烟台 264025;海军航空工程学院兵器科学与技术系, 烟台 264001)

摘 要
提出一种基于张量代数的核主成分分析方法来进行特征提取。该方法可以有效避免维数过高导致计算消耗过大,并合理利用已知训练样本的类别信息。算法先对每一类目标使用核主成分分析手段以形成其各自的特征空间;再通过张量积将所有的特征映射到一高维线性空间;随后直接在此空间上进行线性的主成分分析,即可构造出了适宜的特征空间。其既能有效反映各类样本特征,又能比直接使用核主成分的方法极大降低计算所需的消耗。目标识别实验表明,该方法与直接使用核主成分方法构造特征空间的方法进行比较,在保持识别效果的前提下,可以明显降低计算的消耗与存储的需求。
关键词
SAR images recognition based on kernel principal component analysis

Yu Hongyun,Jiang Tao,Guan Jian(Research Institute of Information Fusion, Naval Aeronautical Engineering Institute, Yantai 264001, China; School of Mathematics and Information Ludong University, Yantai 264025, China;Department of Ordnance Science and Technology Naval Aeronautical Engineering Institute, Yantai 264001, China)

Abstract
A kernel principle component analysis method based on tensor algebra is proposed for feature extraction.It can reduce the huge computation cost due to increasing dimensions,while considering the information of known classes.First the kernel principle component analysis method is applied to each class of targets to build their corresponding feature spaces.Then,the collection of feature spaces is unified into a higher dimensional space after introducing the operation of the tensor product.Hence,a linear principle component analysis method can be directly applied on this feature space in order to construct the proper feature space to both reflect the characters of each class and lower the cost of computation.The recognition experiments showed that the cost of computation and memory can be decreased heavily compared to the approach that builds the feature space by using the kernel principle component analysis method directly.
Keywords
QQ在线


订阅号|日报