测绘学报

• 学术论文 • 上一篇    下一篇

基于再生核Hilbert空间的小波核函数支持向量机的高光谱遥感影像分类

谭琨,杜培军   

  1. 中国矿业大学环境与测绘学院地理信息与遥感科学系
  • 收稿日期:2010-01-06 修回日期:2010-05-28 出版日期:2011-04-25 发布日期:2011-04-25
  • 通讯作者: 谭琨

Wavelet Support Vector Machines based on reproducing kernel Hilbert space for Hyperspectral Remote Sensing Image Classification

Kun Tan   

  • Received:2010-01-06 Revised:2010-05-28 Online:2011-04-25 Published:2011-04-25
  • Contact: Kun Tan

摘要: 本文在研究再生核Hilbert空间的支持向量机和小波分析理论的基础上,构建了基于再生核Hilbert空间的小波核函数支持向量机(小波支持向量机)。在高光谱遥感分类方面,支持向量机存在着分类精度不高和参数选择困难等问题。文中提出了一种再生核Hilbert空间的小波核,其是一种多维小波函数,可以逼近任意非线性函数,有效的解决了参数估计的影响。试验中选取了北京昌平地区的国产高光谱数据(Operational Modular Imaging Spectrometer II (OMIS II))以及国外的ROSIS传感器获取的意大利Pavis大学的数据,分析了小波支持向量机的分类性能和分类精度。在本文的试验中,小波支持向量机在应用小波核函数Coiflet的时候能达到最高的分类精度。与传统的分类方法进行比较,小波支持向量机比光谱角制图分类器和最小距离分类器能达到更高的分类精度,同样比常用的径向基核函数具有更好的分类精度。小波支持向量机是一种新的高光谱遥感分类方法,能得到很好的应用。

Abstract: Studying on the Support Vector Machine (SVM) theory based on reproducing kernel Hibert Space and the wavelet analysis, I constructed the wavelet SVM (WSVM) classifier based on wavelet kernel fuctions. SVM applied hyperspectral classification exists a bottleneck and kernel parameters selection. The wavelet kernel in RKHS is a kind of multidimensional wavelet function that can approximate arbitrary nonlinear functions. Implications on semiparametric estimation are proposed in this paper. By experimented the hyperspectral image with the 64 bands Operational Modular Imaging Spectrometer II (OMIS II) data of Changping Area, Beijing City and ROSIS data of the center of university of Pavia, the classifiers performance and accuracy of WSVM were obtained. In my experiments, the WSVM classifier was demonstrated to be most accurate when it used Coiflet Kernel function of wavelet analysis. Compared with some traditional classifiers (Spectral Angle Mapping classification (SAM) and Minimum Distance classification (MDC)) and classic kernel (Radial Basis Function kernel) of SVM, it indicated that wavelet kernel SVM classifier had the most accurate. Use of the WSVM classifier is a novel approach which improves the accuracy of hyperspectral image classification and expands the possibilities for scientific interpretation and application.