An improved composite kernel framework for hyperspectral image classification using canonical correlation analysis

摘要

This letter presents an improved method of composite kernel framework for hyperspectral image (HSI) classification where we apply canonical correlation analysis (CCA) in the process of combining the spectral and spatial information. This method exploits the potential correlation between the spectral and spatial information since the latter is always extracted from the former. To demonstrate a good performance of the proposed method, support vector machine (SVM) is adopted for evaluation purposes. Experiments on two real HSIs datasets demonstrate: 1) enhanced classification accuracy and robustness compared to the approaches without CCA; and 2) a low cost of calculation since the employment of CCA reduces dimensions effectively.