Abstract: The large amount of data for remote sensing image,more band number and more information redundancy bring difficulties to the further interpretation of the image.To solve this problem,using mutual information between adjacent bands and correlation coefficient matrix between the entire band to group the band,and using the band index and spectral angle mapping algorithm,band selection method for a specific region of interest was proposed.First corrected all of the valid band was grouped and divided in to subspace,and then the index of the biggest bands in the various sub-space was removed.Finally,based on surface features the spectral separability select the best band combination.The trial and the comparision with the common band selection method compares show that the method proposed in this paper has the obvious object extraction effect.

Key words: Hyperspectral image,Mutual information,Correlation coefficient,Band index,Spectral angle mapping