The invention relates to a waveband selection method based on a chaotic binary gravitational search algorithm. The waveband selection problem of a hyperspectral image is optimized and solved by using the chaotic binary gravitational search algorithm so that the superior waveband subsets of the hyperspectral image waveband can be rapidly acquired, and the method can be used for the technical field related to image processing and pattern recognition. The high-quality feasible solution of the waveband selection problem can be found in the acceptable time cost without manual specifying of the waveband dimension required to be selected so that the great balance can be intelligently obtained between the correct recognition rate and the waveband dimension and the superior waveband subsets can be searched. Waveband selection of the original waveband data sets of the hyperspectral image is performed by using the chaotic binary gravitational search algorithm so that the irrelevant or redundant wavebands can be eliminated, the wavebands having high influence on the classification correct rate are taken out, the classification computing time for the irrelevant or redundant wavebands can be reduced and the correct rate and the efficiency of image classification can be further enhanced.