The method of selecting the best classification bands from hyperspectral images based on genetic algorithm and rough set

Selecting the best classification bands from hyperspectral images for particular remote sensing application is one of the most important problems in utilizing hyperspectral images. In this paper, the best classification bands selection problem is regarded as optimal feature subset selection problem and the bands in original bands set are divided into redundant and irrelevant. In order to eliminate these two type bands, a multi-level optimal classification bands selection model from hyperspectral images based on genetic algorithm and rough set theory is proposed. Through the initial two steps of the multi-level model, the dimension reduction step and the genetic algorithm based filter step, most of redundant and irrelevant bands are deleted from the original images bands set. From the machine learning perspective, the multi-level model can take the both advantages of the filter and wrapper models.