Latent Low-Rank Projection Learning with Graph Regularization for Feature Extraction of Hyperspectral Images

Due to the great benefit of rich spectral information, hyperspectral images (HSIs) have been successfully applied in many fields. However, some problems concern also limit their further applications, such as high dimension and expensive labeling. To address these issues, an unsupervised latent low-rank projection learning with graph regularization (LatLRPL) method is presented for feature extraction classification HSIs this paper, which discriminative features can be extracted from view space by decomposing matrix into two different matrices, benefiting preservation intrinsic subspace structures regularization. Different embedding-based methods that need phases obtain low-dimensional projections, one step enough LatLRPL constructing integrated model, reducing complexity simultaneously improving robustness. improve performance, a simple but effective strategy exploited conducting local weighted average on pixels sliding window HSIs. Experiments Indian Pines Pavia University datasets demonstrate superiority proposed method.