用简化核主成分分析法实现高光谱遥感影像降维

摘要

The paper employed a novel method based on Nystrom algorithm to realize dimensionality reduction of hy-perspectral remote sensing image. First, part of the samples was extracted randomly to form sub kernel matrix whose eigenvectors were computed. Then the process above was iterated to compute the new kernel and update the eigenvectors. Finally the image after dimensionality reduction was produced with the last eigenvectors. This method was prepared with KPCA in time consumption, the quantity of extraction feature information and classification effect with datasets OMIS and ROSIS employed. Experimental results show that with contrast to KPCA, SKPCA ( Simplified KPCA ) had comparative performance in feature extraction and classification effects but apparently higher computing speed.%提出用基于Nystr(ǒ)m算法的简化核主成分分析方法(SKPCA)实现高光谱遥感影像的快速降维.首先随机选取部分样本构成子核矩阵并计算其特征向量,然后进行矩阵外推迭代得到近似核矩阵,并分解近似核矩阵不断更新特征向量,最后实现高光谱影像的降维处理.利用OMIS与ROSIS遥感影像进行试验,从运算速度、提取特征信息量以及分类后效果对SKPCA和KPCA(未简化的核主成分分析法)进行比较,结果表明,SKPCA和KPCA提取的特征信息量相当,提取特征与分类效果相近,但SKPCA的运算速度至少要高于KPCA数百倍.