Abstract: In order to solve the dimension reduction problem of hyperspectral image to improve the classification algorithm’s classification accuracy rate and the problem that hyperspectral image usually contains little labeled samples,we proposed a hyperspectral image algorithm based on a semi-supervised neighborhood preserving embedding algorithm and improved k-Nearest Neighborhood classifier．This algorithm uses both the labeled samples and the unlabeled samples of the neighborhood based on Neighborhood Preserving Embedding to get the neighborhood embedding structure,and improve the classification feature through raising weight of the labeled neighboring samples,and thus improving the sample accuracy rate of KNN classifier．The experimental results on the Urban and Indian Pine data sets show that the accuracy rate of the proposed method is improved by more than about 8.7%,3.6%,respectively,and thus the classification performance has been improved clearly.

Key words: Hyperspectral image classification,Dimension reduction,Neighborhood preserving embedding,Semi-supervised learning