应用PCA和ICA法对岩心光谱的定量解译

The application of PCA and ICA methods in quantitative interpretation of mineral hyper-spectrum

岩心光谱属于混合像元光谱,维度多、数据量大,在端元数量、端元光谱及混合矩阵未知的情况下,定量解译岩心光谱以提取岩心所含矿物信息难度大,噪声的存在使问题更加复杂。文章应用PCA和ICA法定量解译岩心光谱主要有三步:采用PCA法预处理混合像元光谱矩阵,在新的特征空间中保留特征值较大的少量特征矢量,有效滤除能量较小的成分,只保留主要成分信息,同时滤除数据中的噪声;采用ICA法分离岩心混合像元,得到混合像元中的端元光谱集,通过矿物识别获取岩心矿物成分;对端元光谱进行归一化处理,基于线性光谱混合模型进行丰度反演,得到岩心混合像元中各端元丰度。通过研究仿真数据获得光谱定量解译的一般规律,创建有效的算法模型,再处理实际测量的岩心光谱,获得了比较理想的结果。

The core hyper-spectrum is of mixed pixels with multi-dimensions and massive data.The quantitative interpretation of hyper-spectrum is extremely difficult when the end number of pixel,end spectrum and the mixing matrix are all unknown.If noise appears it is much complicated.with the application of PCA and ICA methods a three-step process is developed for the quantitative interpretation.Firstly PCA method is used to preprocess the pixel spectrum matrix and in new characteristic space several characteristic vectors with larger eigenvalues are kept and simultaneously noise filtered,Secondly,ICA method is used to un-mix the pixels,obtaining collection of end spectrum and identifying mineral components in the core.Finally,the end spectrum is normalized and abundance inversion carried out and each end abundance obtained.The quantitative interpretation law of hyper-spectrum is resulted from study on the simulation data.Algorithm model is built based on the law and the measured spectral of core processed by the model with ideal results.