光谱相似性测度是高光谱遥感影像定量化分析和精细地物直接识别的基础, 光谱特征的选择和刻画方式是光谱相似性测度的关键。 研究表明, 利用光谱的单一特征无法全面反应地物光谱间的相似性, 光谱识别时需要综合考虑光谱的多种特征。 本文在几何距离、 相关系数和相对熵的基础上提出了一种结合多种光谱特征的新型光谱相似性测度, 即光谱泛相似测度（spectral pan-similarity measure, SPM）。 在进行光谱相似性分析时, SPM综合考虑光谱矢量大小、 光谱曲线形状和光谱信息量三种光谱特征信息。 基于美国地质调查局矿物光谱库和机载OMIS高光谱遥感影像进行试验, 结果表明相对只考虑一种或两种特征的光谱相似性测度, SPM具有更强的光谱判别能力和更小的光谱识别不确定性。

Spectral characterization and feature selection is the key to spectral similarity measure which is the basis of both quantitative analysis and accurate object identification for hyperspectral remote sensing. However, spectral similarity measures using only one spectral feature are usually ambiguous in their distinction of similarity. Multiple spectral features integration is needed for objective spectral discrimination. We present a new spectral similarity measure, Spectral Pan-similarity Measure (SPM), based on geometric distance, correlation coefficient and relative entropy. Spectral Pan-similarity Measure objectively quantifies differences between spectra in three spectral features, the vector magnitude, spectral curve shape and spectral information content. The performance of different spectral similarity measures is compared using USGS Mineral Spectral Library and real (i.e., Operational Modular Imaging Spectrometer, OMIS) hyperspectral image. The experimental results demonstrate that the new spectral similarity measure is more effective than the spectral similarity measure taking into account only one or two features both in spectral discriminatory power and spectral identification uncertainty.