中文摘要：

从月壤光谱中挖掘矿物含量信息是月球遥感探测的重要内容之一，矿物光谱混合规律复杂，光谱特征对测试环境、粒度、地形、背景极为敏感是难点。提出并试验了一种基于光谱分解反演矿物含量的方法。在光谱分解前采用光谱连续统去除突出光谱谱形，分解中综合运用了光谱反射峰（介于两个吸收谷之间的光谱曲线向上凸起的部分）与吸收谷，分解后采用Hapke模型修正光谱非线性混合的影响，在混合端元光谱、端元化学成分与粒度未知情况下，实现了对月壤四种主要矿物橄榄石、斜方辉石、单斜辉石、斜长石的43条混合光谱的盲分解与矿物含量反演，反演的残差、误差、线性拟合相关系数（反演含量与真实含量）均值分别为5．0Vol％，14．4V01％，0．92。该方法与反演结果可为月球高光谱矿物识别与填图提供参考与依据。

英文摘要：

Estimating minerals abundance from reflectance spectra is one of the fundamental goals of remote sensing lunar explo- ration, and the main difficulties are the complicated mixing law of minerals spectrum and spectral features being sensitive to sev- eral kinds of factors such as topography, particle size and roughness etc. A method based on spectral unmixing was put forward and tested in the present paper. Before spectra are unmixed the spectral continuum is removed for clarifying and strengthening spectral features. The absorption features and reflectance features （the upward curving parts of spectra between absorption fea- tures） are integrated for unmixing to improve the unmixing performance. The Hapke model was used to correct unmixing error due to nonlinear mixing of minerals spectra. Forty three mixed spectra of olivine, clinopyroxene, hypersthene and plagioclase were used to validate the above method. The four minerals abundance was estimated under the conditions of being unaware of endmember spectra used to mix, granularity and chemical composition of minerals. Residual error, abundance error and correla- tion coefficient between retrieved and true abundance were 5.0 Vol%, 14. 4 Vol% and 0. 92 respectively. The method and result of this paper could be referred in the lunar minerals mapping of imaging spectrometer data such as M3.