Collaborative Representation Ensemble Using Bagging for Hyperspectral Image Classification

Collaborative representation classification (CRC) is sensitive to the regularization parameter. In order to overcome this drawback, we propose a novel Bagging-based collaborative representation classification (Bags CRC) for hyperspectral image classification, which combines CRC and Bagging together. The principal idea of Bags CRC is to generate diverse CRC classification results using bootstrape sample method, which can enhance single classifier accuracy and diversity simultaneously. In addition, tangent collaborative representation classification (TCRC) has demonstrated its better performance than that of CRC. Here, TCRC is adopted as base classifier in Bagging framework, then the Bagging-based TCRC (Bags TCRC) is proposed. The effectiveness of the proposed method is investigated on two real hyperspectral data sets. The experimental results show that both Bags CRC and Bags TCRC outperform their single classifier counterpart, respectively. Bags TCRC provides the superior performance compare with Bags CRC and random forest.