基于实测高光谱数据空间特征的光谱混合模型

A Spectral Mixture Model Based on Spectral Spatial Character of Measured Hyperspectral Data

为提高高光谱混合像元分解精度,利用地物多角度二向性反射平台和ASD FieldSpec3Hi-Res便携式地物波谱仪,设计等距离/等面积实验,考虑探测距离远近对混合光谱的影响,获取不同覆盖条件下叶片与方解石的混合光谱,找出光谱数据空间特征变化规律,提出等距离/等面积模型来消除空间位置对混合光谱的影响,并将模拟混合光谱与实测混合光谱进行对比.通过实测数据分析,混合反射率分布的权重系数随探测单位面积点与探头距离呈高斯变化规律;与线性模型和线性改进模型进行光谱混合模拟结果相比,应用权重系数高斯分布规律以等距离/等面积模型进行混合光谱模拟,其模拟结果相似度平均增加了1.20%,均方根误差平均降低了7.78%.等距离/等面积光谱混合模型考虑了光谱空间变化特征,提高了光谱混合模拟的精度,为进一步研究高光谱数据混合像元分解提供了新的方法.

In order to enhance the estimation of mixed spectral model, the equidistant/homalographic model is established to analyze the spectral spatial character to simulate the mixture spectra. Based on the reflex platform and FieldSpec 3 Hi-Res port- able spectrum instrument, the equidistant/homalographic experiment, which takes the effect of distance between optical fiber probe and detected endmember into account, was designed to acquire the mixed spectral reflectance of calcite and green leaf. The measured mixed spectra analysis shows the weight coefficients of distribution change with the distance between the detec- ted endmember and the probe is in a Gauss distribution. Compared with the linear spectral mixture model and improved linear spectral model, the results simulated by the equidistant/homalographic model is 1.20% greater in similarity and 7.78% lower in RMSE. Considering the influence exerted by spectral spatial structure on mixed spectral simulation, the equidistant/homalographic model proves to improve the accuracy of mixed spectral simulation and a new method for unmixing the mixed pixel of hyperspectral data.