Research On Endmember Extraction Algorithm And Application Of Hyperspectral Remote Sensing Image

The principle of the hyperspectral remote sensing technology is obtaining target information about electromagnetic spectrum in the UV, visible, near-infrared and mid-infrared region using the spectroscopic techniques, and showing as a cube in the space dimension and spectroscopy dimension. The hyperspectral remote sensing has the characteristics of many bands, much higher spectral resolution, and the amount of data. There exits several reasons, like a single element with a variety of surface features, the atmosphere during transmission and remote sensing instrument itself, resulting in a large numbers of mixed pixels in the image. The existence of mixed pixels directly influences the classification and the identification of the hyperspectral image. Therefore, the unmixing of the mixed pixels becomes a hot issue to be solved in the development of remote sensing technologyIn this paper, the research is focus on the endmember extraction of the hyperspectral remote sensing image based on linear spectral decomposition model, and proposing two improved unmixing methods. The main research work and achievements are as follows:(1) Because of hyperspectral remote sensing image with the features of a large amount of data and its complex computational, the paper proposes a new endmember extraction method of N-FINDR based on ICA, and through the experiment comparing with the method of N-FINDR based on MNF, it proves new method is effective according to covariance analysis.(2) Compared to the traditional endmember extraction using the spectrum information, and AMEE with tedious endmember discrimination using spatial information and spectral information, a method of AMEE based on the maximum distance method with coordinate has been proposed, and through the comparative experiment, the covariance of abundance error image has been greatly improved, and it proves new method is feasibility.