基于Bhattacharyya距离的典型地物波谱特征差异性分析

The Spectral Characteristics Separability Analysis of Spectral Database of Typical Objects of Land Surface based on Bhattacharyya Distance

地物波谱数据主要应用于定量遥感与影像分类等相关基础研究,对各条光谱曲线之间进行定量化的光谱差异性分析具有重要意义。从USGS及JHU地物波谱库中挑选了在土地覆盖分类层次具有意义的植被(73条)、人工材料(100条)与土壤(30条)3种类型共203条地物波谱数据,以分层分类体系在4.2～2.5μm的波长范围内分析比较各类典型地物材料的光谱特征,以B距离(Bhattacharyya Distance)作为指标定量计算不同类别地物波谱间的光谱差异性。结果表明:波谱库中金属、砖石和混凝土3类人工材料光谱对于植被、土壤等自然材料光谱具有较大的光谱差异性,而塑料与自然地物间的光谱差异度最小,在此基础上统计了最能反映这些地物光谱特征差异的最优波段。该方法能够量化多种光谱曲线间的差异性并得到最佳的区分波段,从而为地物材料光谱及高光谱数据分类提供参考。

Typical land surface spectrum data are mainly applied in related basic research of quantitative re mote sensing and image classification. We selected representatively spectral library set of vegetation spec trum（73 items） ,manmade spectrum（100 items）and soil spectrum（30 items） from USGS and JHU spectral library, and analysed typical separability features of materials spectral features in 4.2 ~ 2.5 ~m wavelength range within a hierarchical classification scheme, Application of Bhattacharyya distance to quantitatively calculated among different categories objects spectrum spectral differences, the calculation results show that spectrum separability metal, brick and concrete manmade material spectrum for vegetation, soil and other natural materials spectrum have greater spectral differences, and separability of plastic between natu ral features is smaller. Additionally, an evaluation of the most suitable wavelengths for separation of spec tral library set identified specific spectral features that provided the best separation. Based on the statistical characteristics of spectral differences could reflect the optimal band. The study provides a basic knowledge reference of spectral discrimination analysis in a variety of material spectrum and also have the certain ref erence significance of remote sensing image land classification in a larger scale.