Abstract: Without using the data category information,Locality Preserving Projection algorithm is an unsupervised algorithm in the dimension reduction of hyperspectral images,so it has not a very good performance in the aspect of feature extraction．To resolve this problem,an algorithm based on semi-supervised locality preserving projection (SSLPP) is proposed in this paper．A graph G_i is constructed by SSLPP combined with the information of unlabeled samples and labeled samples whose are treated in different ways,therefore increasing the weights of congener samples and being beneficial to feature extraction．Experiments on the AVIRIS KSC set and Botswana data set show that the algorithm proposed in the paper can find the high dimensional space data intrinsic structure,effectively improving the overall accuracy of the classification.

Key words: Hyperspectral images,Dimension reduction,Semi-supervised learning,Semi-supervised locality preserving projection