Semisupervised Feature Extraction Based on Collaborative Label Propagation for Hyperspectral Images

This letter presents a semisupervised feature extraction based on collaborative label propagation (SSCLP) for hyperspectral images (HSIs). SSCLP first proposes a novel collaborative label propagation method to predict the labels of unlabeled data that are termed weak labels. Then, SSCLP combines the known labels and the predicted weak labels to construct two new discriminative matrices. Finally, the discriminative matrices are utilized to find an optimal transformation matrix to achieve feature extraction for HSIs. The proposed SSCLP not only preserves the compactness of intraclass and the separability of interclass but also explores the weak labels information and the local neighbor information of unlabeled data. Experiments on the Pavia University and Kennedy Space Center datasets demonstrate that the proposed SSCLP has a better performance than other related methods.