光谱特征提取对近似矿物光谱差异性影响分析

Research on the Impact of Absorption Feature Extraction on Spectral Difference Between Similar Minerals

矿物光谱特征是基于光学遥感数据对矿物进行种类识别及定量反演的理论基础,光谱特征提取是高光谱数据常用的技术手段,但在多光谱数据中较少涉及。近似矿物识别是矿物光谱分类应用中的难点,目前还缺少有效指标来指示近似矿物类别光谱的差异性。光谱特征提取有望提高矿物分类精度,但该处理对近似矿物光谱差异性的影响还缺少相关研究。本文从矿物光谱差异性的原理出发,通过类间和类内光谱角的比值体现不同类别群体差异,并引入样本量因素,提出了类别可分比作为近似矿物光谱差异性的指标。以明矾石和高岭石两种近似矿物为例,对USGS光谱库光谱及Hyperion,ASTER,OLI等传感器的模拟数据进行光谱特征提取处理,通过对比处理前后矿物光谱差异性的变化,分析光谱特征提取对近似矿物光谱差异性的影响。实验结果表明,有效的光谱特征提取可以显著提高近似矿物光谱差异性,并且光谱分辨率越高,近似矿物光谱差异性越大。此外,光谱分辨率及中心波长设置对于包络线去除结果有很大影响,多光谱数据吸收特征提取效果有待进一步提高。该研究为今后近似矿物光谱识别精度的提高奠定了基础,也为未来新型遥感找矿传感器参数设置提供了参考。

Diagnostic absorption features can indicate the existence of specific materials,which is the foundation of mineral analysis with optical remote sensing data.In hyperspectral data processing,the most commonly used method to extract absorption feature,is Continuum Removal（CR）.As for multispectral data,Principle Component Analysis and other indirect methods were used to extract absorption information,and little research has been done on full-band absorption feature extraction.Classification of similar minerals is one of the major difficulties in mineral spectral analysis,while there is no valid index for spectral difference between similar mineral groups.Absorption feature extraction may improve the classification accuracy,but there is no research to investigate the impact of absorption feature extraction on spectral difference between similar minerals.This paper summarized the principle of mineral spectral difference,and proposed the concept of Class Separability Ratio（CSR）,which was verified to be a valid index for spectral difference between similar mineral categories.Through comparison experiments on alunite and kaolinite spectra,including USGS spectral library spectra and resampled spectra in accordance with the band settings of HYPERION,ASTER and OLI,the impact of absorption feature extraction on spectral difference between similar minerals were investigated.Experimental results show that valid absorption feature extraction can greatly enhance the spectral difference between similar minerals,and the spectral difference is positively correlated with spectral resolution.Besides,the results of CR can be severely affected by spectral resolution and band center positions,and the absorption feature spectra extraction results for multispectral datasets need to be improved.This research laid the foundation of precise identification between similar mineral categories,and provided valuable reference for the band settings of future geology remote sensing sensors.