Acta Geodaetica et Cartographica Sinica ›› 2015, Vol. 44 ›› Issue (11): 1227-1234.doi: 10.11947/j.AGCS.2015.20140600

Previous Articles     Next Articles

Informative Vector Machine Classification for Hyperspectral Imagery

TAN Xiong, YU Xuchu, QIN Zhiyuan, ZHANG Pengqiang, WEI Xiangpo   

  1. Institute of Surveying and Mapping, Information Engineering University, Zhengzhou 450001, ChinaAbstract
  • Received:2014-11-18 Revised:2015-03-10 Online:2015-11-20 Published:2015-11-25
  • Supported by:
    The National Natural Science Foundation of China(Youth Science Foundation)(Nos. 41201477;41401534), The Open Fund of State Key Laboratory of Geographic Information Engineering(No. SKLGIE2013-M-3-1) The Scientific Research Foundation for Public Welfare Industry of Surveying and Mapping and Geographic Information(No. 201412007)

Abstract: Informative vector machine is a method of sparse Gaussian process based on Bayesian theory, which has high speed in model training, small consuming in memory, strong effective in sparseness and good forecasting performance. In this paper, the Gaussian process regression model is introduced firstly, and then a hyperspectral imagery classification method based on informative vector machine is brought forward. Secondly, to solve the problem of non-Gaussian noise model in the Gaussian process classification, the classification problem is transformed into a regression problem by using the assume density filtering algorithm, after which model is trained by maximizing the marginal likelihood function. Finally, the number of informative vector is chosen in active subset to achieve the purpose of sparse. According to the experimental results of ROSIS images, the advantages of hyperspectral imagery classification method based on informative vector machine are validated.

Key words: hyperspectral imagery, informative vector machine, classification

CLC Number: