Partition Optimal Band Selection Method for Hyperspectral Image

Abstract

Hyperspectral data has the characteristics of large number of bands, narrow band width and large amount of data, which brings great difficulties to the further interpretation of the image. The partition optimal band selection(POBS) is proposed to select a certain number of bands according to the actual needs. The weight calculation formula of the band includes three parameters: standard deviation, information entropy and correlation coefficient, so it can be guaranteed that the selected band subset contains a large amount of information and low correlation. In addition, the whole band space is evenly divided into several band subspaces, and then the band with the largest weight is selected in each band subspace. The bands can be more dispersed through the uniform partition of the band space. It can ensure that the bands in the band subset are more representative. The classification test is carried out on two hyperspectral datasets by KNN method, and the classification results show that POBS selects a large amount of band information, and the divisibility of the ground object is good.