北京航空航天大学学报 ›› 2013, Vol. 39 ›› Issue (5): 635-639.

• 论文 • 上一篇    下一篇

基于光谱特征的自适应子空间波段选择方法

徐秋, 李娜, 赵慧洁, 李旭东   

  1. 北京航空航天大学 精密光机电一体化技术教育部重点实验室, 北京 100191
  • 收稿日期:2012-05-16 修回日期:2013-05-07 出版日期:2013-05-31 发布日期:2013-05-22
  • 基金资助:
    国家自然科学基金资助项目(61008047,61177008); 长江学者和创新团队发展计划资助项目(IRT0705); 中国地质调查局资助项目(1212011120227)

Adaptive subspace band selection method based on spectrum characteristics

Xu Qiu, Li Na, Zhao Huijie, Li Xudong   

  1. Key Laboratory of Precision Opto-mechatronics Technology, Ministry of Education, Beijing University of Aeronautics and Astronautics, Beijing 100191, China
  • Received:2012-05-16 Revised:2013-05-07 Online:2013-05-31 Published:2013-05-22

摘要: 针对现有波段选择方法无法依据感兴趣地物划分数据子空间及子空间划分容易受背景地物影响等问题,提出一种基于光谱特征的自适应子空间波段选择方法.利用感兴趣地物的光谱曲线,建立光谱自适应因子(SAF,Spectral Adaptive Factor);通过聚类划分子空间,在每个子空间内,利用J-M距离筛选类别可分性最大波段作为子空间最优波段,从而得到整个数据空间的最佳波段组合.利用AOTF (Acousto-Optic Tunable Filter)成像光谱仪数据与波段指数(BI,Band Index)和基于类间可分性最佳波段选择两种方法比较,SAF方法选择的最佳波段组合,包含信息量丰富,地物之间差异较大,且地物平均J-M距离大于另外两种方法.用最大似然法(maximum likelihood)对最佳波段组合进行分类,总体分类精度为96.8%,Kappa系数为0.894.结果验证了SAF方法的有效性和实用性.

Abstract: Adaptive subspace band selection method based on spectrum characteristics was proposed to solve the problems, including the existing methods which couldn't divide subspace by studied features and background features easily affected the result of subspace division.Spectral adaptive factor (SAF) was established with the spectral curves of studied features, and the whole data space was divided into some subspace based on clustering. In each subspace, Jeffreys-Matusita distance was calculated to choose the maximum class separability band as the optimal band. The optimal bands combination was achieved.With the data of acousto-optic tunable filter (AOTF) imaging spectrometer,an experiment was accomplished to compare with other band selection methods, involving band index (BI) method and the optimal bands selection method based on the classes distinguish ability. Experimental results show that the optimal bands combination of the proposed method contains better performance and studied features are shown more significant difference. And the average of Jeffreys-Matusitadistance of all classes of the proposed method is the greatest of all methods.Maximum likelihood classification method was also implemented on the images of the optimal bands combination of the proposed method. As a result, the overall accuracy is 96.8% and Kappa coefficient is 0.89.The experiment indicates that the proposed method is effectiveness and practicability.

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发