Material Identification in Hyper-spectrum Remote Sensing Images

A hyper-spectrum image contains a very large quantity of information measured at a geographic region and over hundreds of frequencies. Such images can lead to interesting applications in earth and environment protection research. An essential topic in hyper-spectrum imaging is how to identify materials, or their spectra, which are present in a region of the earth's surface. In this paper, this topic will be considered through the “unmixing” problem in which spectrum or material identification from a mixture of spectra is carried out at each pixel in a hyper-spectrum image. The spatial constraint among neighboring pixels will be analysed to improve the performance of spectrum identification. The method tested with artificial and real data has shown promising results