Robust joint nearest subspace for hyperspectral image classification

This letter presents a novel hyperspectral image (HSI) classification method based on robust joint nearest subspace and contextual prototype learning. First, we present a robust joint nearest subspace method to solve the HSI classification problem by exploiting a set-to-class distance with robust distance metric to consider both spectral and spatial characteristics effectively. Second, we develop an objective function to learn contextual prototypes robustly and present an iteration algorithm to solve it. Based on the learned contextual prototypes, the HSI classification performance can be further improved. Finally, we conduct numerous experiments to validate the effects of different parameters and components and to compare the proposed method with other algorithms on three popular data sets. The experimental results show that the proposed method achieves better performance than other competing algorithms.