Mixed pixel decomposition of mineral spectrum based on EMD-ICA method

Abstract

Mineral spectrum is a mixed pixel consists of various mineral spectra. The types and contents of the component minerals can be obtained through mixed pixel decomposition based on EMD-ICA method. The independent component analysis (ICA) can decompose the mixed pixels, but requires that the number of mixed pixels should be no less than the number of end members. However, this requirement usually cannot be met in practice, due to limitations on measuring conditions. Besides, it is hard to find out the number of source signals in measurement signals in advance. To remedy this, the empirical mode decomposition (EMD) is used to obtain intrinsic mode functions (IMFs), which can be combined with the mixed pixel spectra as the input data matrix of ICA. With the EMD-ICA method, the estimate of the end member spectra can be separated from a single mixed pixel spectrum. Studies have shown that the higher the spectral content of end members in mixed pixels, the higher the quality of the estimates of end member spectra. Choosing different numbers of IMFs in ICA leads to different numbers of estimates of end member spectra. When the number of the estimates is less than the actual number of end member, the estimates obtained correspond to the end members with higher spectral content. By applying this method in quantitative analysis of mineral spectrum, we can obtain information about the end members it contains.