Discriminative graph-based dimensionality reduction for hyperspectral image classification

A novel discriminative graph-based dimensionality reduction (DGDR) model is proposed for HSI dimensionality reduction and classification. The core idea of the proposed method is to search for a projection function by minimizing the similarity term that contains the relation of within-class scatter and maximizing the dissimilarity term that contains the relation of between-class distance. The proposed method pulls close together samples being similar while pushing those dissimilar samples apart in the projected latent space. The edges of the graphs are measured by kernel. Furthermore, the multiscale DGDR (MS-DGDR) is introduced to utilize the capability of similarity measure of different scales of kernel and avoid finding the optimal scale simultaneously. Experiments are conducted on a real HSI. The corresponding results demonstrate the effectiveness of the proposed method for HSI both in improving classification accuracy with the same and fixed feature dimensionality and feature dimensionality reduction with the same requirement of classification accuracy, compared with several state-of-the-art dimensionality reduction algorithms.