Dimension Reduction for Hyperspectral Image Classification Based on Sparse Spectral Spatial Locality Preserving Projection

As a dimensionality reduction method, local preserving projection (LPP) performs well in many high-dimensional data tasks, but there are some limitations in dimensionality reduction of hyperspectral images. 1) Weak ability to distinguish inter class differences, 2) ignoring global structure, 3) sensitive to redundant bands. Therefore, this article proposes a sparse spectral spatial locality preserving projection method to solve the problems faced by traditional LPP. In this method, gray level co-occurrence matrix, improved sparse subspace clustering, LPP, and weighted mean filtering are combined together to fuse spatial texture features and increase sample heterogeneity, global dimensionality reduction of data, preservation of local data structure, and smoothing of classification results, respectively. Three public experimental datasets indicate that the proposed model is significantly superior to other commonly used extraction methods. In short, the proposed method demonstrates powerful discriminative features and improves the classification performance of images.