基于信息量改进主成分分析的高光谱图像特征提取方法  

Feature Extraction for Hyperspectral Images Based on Principal Component Analysis Improved by Information Quantity

作　　者：任智伟 吴玲达 REN Zhiwei;WU Linda(Science and Technology on Complex Electronic System Simulation Laboratory,Space Engineering University,Beijing 101416,China)

机构地区：[1]航天工程大学复杂电子系统仿真实验室,北京101416

出　　处：《兵器装备工程学报》2018年第7期151-154,共4页Journal of Ordnance Equipment Engineering

基　　金：重点实验室基础研究项目

摘　　要：针对传统PCA降维方法仅依据方差贡献率进行特征提取,无法有效利用光谱图像本身信息的问题,提出一种利用信息量改进PCA的降维方法。本文利用信息量改进PCA方法对原始高光谱图像数据进行降维处理,将降维结果输入最小距离分类器进行分类,计算分类精度,并通过数据集Indiana Pines与Salinas验证。实验结果表明:与未改进PCA降维方法相比,信息量改进PCA降维方法在计算代价仅增加o(n)复杂度的情况下,分类精度有明显提高。For traditional PCA dimension reduction methods, feature extraction is only based on the variance contribution rate, which can not effectively use the information of the spectral image itself, and a dimension reduction method using information quantity to improve PCA is proposed. In this paper, the original hyperspectral image data is reduced dimensionally by the information-augmented PCA method. The dimensionality reduction results are input into the minimum distance classifier for classification and the classification accuracy is calculated. The dataset is validated by the datasets Indiana Pines and Salinas. The experimental results show that, compared with the unmodified PCA dimension reduction method, the information-improving PCA dimension reduction method significantly improves the classification accuracy when the computational cost only increases the o （n） complexity.

关 键 词：高光谱图像 特征提取 主成分分析法 信息量 最小距离分类器 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]