Linear Regression Fisher Discrimination Dictionary Learning for Hyperspectral Image Classification

摘要

In this paper, we propose a novel dictionary learning method for hyperspectral image classification. The proposed method, linear regression Fisher discrimination dictionary learning (LRFDDL), obtains a more discriminative dictionary and a classifier by incorporating linear regression term and the Fisher discrimination into the objective function during training. The linear regression term makes predicted and actual labels as close as possible; while the Fisher discrimination is imposed on the sparse codes so that they have small with-class scatters but large between-class scatters. Experiments show that LRFDDL significantly improves the performances of hyperspectral image classification.