高光谱图像基于像素结构的改进PCA算法

An Improved PCA Algorithm using Pixel-Based Structure in Hyperspectral Imaging

主成分分析法(PCA)作为一种常用的降维算法,被广泛的应用到如高光谱图像处理等需要进行大量数据处理的应用中。PCA的主要目的是利用正交变换,将具有相关性的高维数据的分量转换为线性不相关的新的成分变量,但当矩阵维数超过百万时候会造成严重的计算困难问题。本文针对PCA运算中协方差矩阵计算过程中内存调度的问题,提出了一种基于像素结构的改进的协方差矩阵计算方法,可以在确保与常规PCA具有相同性能的同时有效地降低计算所需的存储器规模。实验中分别采用传统PCA算法和改进算法对高光谱图像数据进行特征提取后利用支持向量机(SVM)进行分类,对比结果验证了改进算法的有效性和可靠性。

Principal component analysis （ PCA）, which convert correlatedhigh dimensional data to linearly uncorrelatedprincipal components by using orthogonal transfannation, is widely used in many applications with requirement o f large dataprocessing, such as hyperspectral imaging. Because o f the computational problem o f PCA which appears on one dimension o f the matrix over 100k, an improved PCA algorithm using pixel-based structure is proposed in this paper. Comparing with traditional PCA, the proposed method can reduce the memory requirement while maintain the preference. Using support vector machines （SVM） to classify both the aew components o f two PCA algorithms, classification results show the proposed algorithm is reliable and effective.