基于FastICA算法的高光谱矿物丰度反演
Hyper-Spectral Mineral Abundance Inversion Based on FastICA Algorithm
Abstract
高光谱分离是定性与定量岩矿分析的核心基础。仿真数据和测量数据可通过fASTICA算法处理,计算混合像元中端元数量及端元光谱,并根据计算结果进行丰度反演。实验结果表明:fASTICA算法即使在端元光谱、混合矩阵未知且没有任何先验知识的情况下,也能有效分离岩矿高光谱混合像元并得到端元的估计光谱,从而实现矿物识别和矿物丰度反演;但当混合像元中参与混合的端元光谱相似度较高或端元光谱的非高斯性较低时,混合像元的分离精度将降低,混合像元的分离精度越高,端元丰度反演结果越准确。 The unmixing of hyper-spectral data is the crucial base of qualitative and quantitative analysis of rock minerals.The simulated or measured hyper-spectral data can be analyzed with the FastICA algorithm to calculate the number of pure-pixels as well as the spectrum of each pure-pixel.Based on the results,the mineral abundance inversion can be carried out.The experiments show that the FastICA algorithm can effectively un-mix the mineral hyper-spectrum and estimate the spectrum of each pure-pixel,even when there is no prior knowledge of the spectrum or the mixing matrix of pure-pixels.This facilitates the mineral identification and the mineral abundance inversion.However,when the spectrums of the pure-pixels are highly similar or the non-gaussianity is bad,the precision of the mineral identification would be reduced,which in turn reduces the precision of mineral abundance inversion and vice versa.