Abstract: Hyperspectral data has the features of multi-band, large amounts of data, etc.For hyperspectral remote sensing image classification, traditional methods spend a long operation time in band selection, and image classification accuracy is not high, so first we use differential evolution (DE) for band selection, effectively reduce the redundancy of information and data dimensions, then show the image based on the result of band selection, sample the typical area of the surface feature to identify, finally use gene expression programming (GEP) to bulid classifier for image classification.In band selection, compared with full search algorithm, DE obtains relative good results in a fast time, in GEP image classification, the classification results superior to the traditional KNN algorithm.