Acta Scientiarum Naturalium Universitatis Pekinensis

Previous Articles     Next Articles

Blind Separation of Component Information from Hyperspectral Data

TAO Xin, FAN Wenjie, XU Xiru   

  1. Institute of RS and GIS of Peking University, Beijing 100871;
  • Received:2007-09-18 Online:2008-11-20 Published:2008-11-20

高光谱数据组分信息的盲分解方法

陶欣,范闻捷,徐希孺   

  1. 北京大学遥感与地理信息系统研究所,北京100871;

Abstract: Blind signal separation (BSS) based on the technique of independent component analysis (ICA) was introduced to the quantitative remote sensing field for mixed pixel unmixing. The scale invariant problem of the classical method was solved and the spectral and weight information of components was synchronously gained from hyperspectral data. The algorithm was further improved in the computer numerical simulation experiments, where the method for choosing best spectral coverage for retrieval was presented. Its robustness was also discussed. It was finally applied on the HYPERION hyperspectral image of the study area in Hengshan county, Shanxi Province, for retrieving the vegetation cover in pixels. The accuracy validation by using SPOT5 image shows the high accuracy of this algorithm.

Key words: mixed pixel, independent component analysis (ICA), hyperspectral, blind signal separation (BSS)

摘要: 将基于独立成分分析(independent component analysis,ICA)技术的盲分解方法(blindsignalseparation,BSS)应用于遥感混合像元的定量分解,解决了幅度不确定性问题,实现了从高光谱数据中同时得到定量的组分光谱信息和组分权重信息。通过数值模拟实验提出了光谱反演区间的选择方法,进一步完善了该算法,且讨论了算法的稳健性。以陕西省横山县为试验区,从HYPERION高光谱影像中反演了各像元的植被覆盖度,并利用SPOT5影像进行了精度验证,结果表明该方法具有较高的精度。

关键词: 混合像元, 独立成分分析(ICA), 高光谱, 盲分解

CLC Number: