第36卷，第5期 2016年5月
光谱学与光谱分析 Spectroscopy and Spectral Analysis
Vol. 36,No.5+pp1592-1597
May,2016
Application of EmpiricalMode Decomposition and Independent
Component Analysis for the Interpretation
ofRock-MineralSpectrum
WUFang',JIANGXi-pingl*,YUHan-wen,XIULian-cun
1. College of Science, Nanjing Agricultural University, Nanjing 210095, China 2. Nanjing Artillery College, Nanjing 211132, China
3. Nanjing Institute of Geology and Mineral Resource, Nanjing210016, China
Abstract Rock-mineral spectrum is a mixture of varied mineral spectra, through which we can obtain informa tion about its components quickly and conveniently without any damage to the sample. Empirical mode decom position (EMD) cannot directly decompose source signals from information of the mixture, and independent component analysis (ICA) requires the number of mixed signals to be no less than the number of source sig nals. Combining these two methods, mixed signals can be decomposed using EMD method to obtain intrinsic mode function (IMF), while certain IMFs together with mixed signals can be used as input data matrix of ICA to obtain the source signals. This method overcomes the shortcomings of IMF and ICA. Studies have shown that, the higher content of source signals contained in the mixed signal, the better estimation can be obtained through EMD and ICA. The number of IMFs that participate in ICA decomposition determines the number of approximation of source signals. The accuracy of source signal estimation increases with the correlation coeffi-cient between IMF and mixed signals. By applying this method to quantitative analysis of rock-mineral spec trum, information of the component minerals in rock-mineral can be obtained, which improves the efficiency of component analysis in detecting rock-minerals outside.
KeywordsEmpirical mode decomposition; Independent component analysis; Rock-mineral spectrum
中图分类号：0657.3
Introduction
文献标识码：A
D0I: 10. 3964/j. issn. 10000593(2016)051592-06
of mixed pixels is fundamental to the interpretation of mixed pixel information. It is of great importance in resource inves tigations and geological prospecting to find fast and accurate
Rock-mineral spectra obtained from spectrum scanner features high spectral resolution and abundant information. Since rock-mineral has a variety of components, its rock-min eral spectrum is a mixture of various mineral spectra. The rock-mineral spectrum is known as "mixed pixel", while the mineral spectra are called end-members. Mixed pixels are very common in all types of remote sensing hyper-spectral ima-ges: and synthetic aperture radar imagest?]. Decomposition
Received : 2015-01-02; accepted: 2015-04-18
decomposition methods to obtain the types and contents of the minerals according to the features of rock-mineral spec truma].
In practice, most of measurement signals are mixed sig nals from various source signals, which are non-stationary and nonlinear. Empirical Mode Decomposition (EMD) and Inde pendent Component Analysis (ICA) are two analytical tech niques for non-stationary and nonlinear signals proposed in re
Foundation item: National Key Scientific Instrument and Equipment Development Project (China) (2012YQ050250), Fundamental Research
Funds of Nanjing Agricultural University (KYZ201425)
Biography : WU Fang, (1974—), female, a lecturer in College of Science, Nanjing Agricultural University
Corresponding author
e-mail : jiangxp@njau. edu. cn
e-mail: wufang318@njau. edu. cn