基于优化ICA算法的高光谱矿物识别

混合像元是影响岩矿高光谱由定性解译向定量反演发展的关键因素之一.以往分离岩矿高光谱混合像元,需要先获得参与混合的端元数量及端元光谱,这在许多场合是难以做到的.独立成分分析可以在端元光谱、混合矩阵未知且没有任何先验知识的情况下,有效分离岩矿高光谱混合像元,只要端元光谱是非高斯性信号且满足统计独立性.它实现了矿物识别,并为矿物丰度反演及成分识别打下了基础.通过引入调整因子,改善独立成分分析(Independent Component Analysis,ICA)算法的收敛性.当参与混合的端元光谱相似度较高或者端元光谱的非高斯性较低时,岩矿高光谱混合像元的分离精度将受到影响.

Hyper-spectral Mineral Identification Based on Improved ICA Algorithm

The mixed-pixel problem is one of the major gaps between qualitative interpretation and quantitative inversion of the mineral hyper-spectrum.In previous studies on the unmixing of hyper-spectral data,the prerequisites are the number of involved pure-pixels and their spectrums,which are often unknown in practice.When there is no prior knowledge of the spectrum and the hybrid matrix of pure-pixels,the"independent component analysis"can effectively unmix the mineral hyper-spectrum,as long as the spectrum signals of pure-pixels are non-Gaussian and statistically independent.With the adjustment factors regarding the iterative un-mixing matrix,the improved ICA algorithm has better convergence property.This achieves the mineral identification,which is the foundation of mineral abundance inversion and component analysis.However,when the spectrums of the pure-pixels are highly similar or the non-gaussianity is bad,the precision of the algorithm would be affected.