作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2007, Vol. 33 ›› Issue (09): 204-206.

• 人工智能及识别技术 • 上一篇    下一篇

基于混合概率PCA模型高光谱图像本征维数确定

普 鑫   

  1. (中国科学院研究生院,北京100072)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-05-05 发布日期:2007-05-05

Intrinsic Dimensionality Determination for Hyperspectral Image Based on Mixture of Probabilistic PCA Model

PU Xin   

  1. (Graduate School, Chinese Academy of Sciences, Beijing 100072)
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-05-05 Published:2007-05-05

摘要: 如何有效实现降维是现代成像光谱仪辨识地物类别的一个难点所在。该文在已知高光谱图像地物类别数的情况下,提出了一种采用混合最小描述长度(MMDL)模型选择准则确定高光谱图像本征维数的方法。该方法在期望最大化算法框架下同时实现混合PPCA降维和聚类,并根据MMDL准则确定数据降维维数,可以得到数据在概率意义下的精确的降维表征。仿真数据和真实数据进行的比较实验表明,该方法能精确地选择数据的本征维数。

关键词: 降维, 本征维数, 混合概率主成分分析, 混合最小描述长度准则, 期望最大化算法

Abstract: An intrinsic dimensionality determination method for hyperspectral image with known class number is proposed, which is based on mixture model of probabilistic PCA. Different from common methods that determine the number of dimensionality reduction by setting the number or by eigenvalue thresholding, the algorithm simultaneously conducts dimensionality reduction and clustering under the frame of EM algorithm; and retrieves the intrinsic dimensionality according to the MMDL principle with probabilistically accurate reduced representation of the data. The method can achieve precise results applied to simulated data and real data.

Key words: Dimensionality reduction, Intrinsic dimensionality, Mixture of probabilistic principal component analysis (PPCA), Principle of mixture minimum description length (MMDL), Expectation maximization(EM) algorithm

中图分类号: