The high dimensions of hyperspectral remote sensing image have caused problems to further processing. In order to solve the above problems

this paper proposes an adaptive band selection (ABS) algorithm of dimensional reduction

which selects high informative and low correlative bands by calculating the index of each band. After calculating and rearranging the index of each band

there are two methods of selecting the final bands: one is to select the bands whose index is bigger than the specified index

another is to select the first n bands whose index is the first n bands. In order to testify the effect of ABS method

Bayesian supervised classification was implemented on the dimensionally reduced image. The results of the classification show that highly informative bands can be selected by ABS method

the classification accuracy of selected bands is 10 4% bigger than the original image

and the computing complication is decreased rapidly too.