Hyper-spectral data processing method

The invention provides a hyper-spectral data processing method. The method comprises the following steps: firstly, determining an end member number; then, using a MLIG algorithm to carry out image end member extraction; finally, using a MVC-MRF algorithm to carry out end member optimization and spectral unmixing. In the method, a restricted linear model is simplified; and end member number determination and an end member extraction algorithm are strictly derived by a mathematical formula. The method is strict in a theory. According to linear independence of an end member, a MLIG end member identification algorithm is provided. Calculation of the end member extraction algorithm is simple and efficiency is high. Large data set application can be satisfied. Finally, a MVC-MRF algorithm is constructed to carry out optimization and spectral unmixing on the extracted end member; image end members of all the linearly independent can be extracted; representativeness is possessed; an unmixing error is almost zero and classification precision is high.