基于改进SPA算法的高效降维方法  

An efficient reduction method based on the improved successive projection algorithm

作　　者：崔颖[1,2] 王雪婷 刘述彬 陆忠军[2] CUI Ying;WANG Xueting;LIU Shubin;LU Zhongjun(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Remote Sensing Technology Center,Heilongjiang Academy of Agricultural Sciences,Harbin 150086,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]黑龙江省农业科学院遥感技术中心,黑龙江哈尔滨150086

出　　处：《应用科技》2018年第5期51-55,共5页Applied Science and Technology

基　　金：国家自然科学基金项目(61675051);教育部博士点基金项目(20132304110007)

摘　　要：为解决高光谱图像数据维数高、冗余信息较多、容易出现Hughes现象等问题,将改进的连续投影算法应用到高光谱图像降维处理中。改进的连续投影算法在原始算法基础上,分别采用峰度值和偏度值对初始波段的选择进行限制,在较短的时间内获得了少量高效的特征波段,提高了分类性能和处理速度。在AVIRIS数据基础上,对本文提出的算法进行实验仿真,分别采用相关向量机(RVM)和支持向量机(SVM)分类器进行分类处理,并与改进的连续投影算法和蒙特卡罗算法的结果进行比较,实验结果表明改进算法的降维性能更好。The hyperspectral image data has high dimensionality and much redundant information, which would causeHughes phenomenon easily. In order to solve this problem, the improved Successive Projection Algorithm （SPA） isapplied to the dimension reduction processing of the hyperspectral image data. Based on the initial algorithm, theimproved SPA limits the choice of the initial band with the kurtosis and the skewness respectively, and obtains a smallnumber of high-efficiency characteristic bands in a short time, which improves the classification performance andprocessing speed. Using the AVIRIS data, the algorithm proposed in this paper is simulated by the relevance vectormachine （RVM） and support vector machine （SVM） classifier respectively. The experimental results show that SPAperforms the Monte Carlo algorithm in improving the dimension reduction performance.

关 键 词：高光谱图像 波段选择 降维 连续投影算法 峰度 偏度 支持向量机 相关向量机 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]