Evaluating the performance of hyperspectral feature selection using quantitative multivariate correlation analysis

Many feature selection methods have been proposed in recent years, but there is little work concerning the evaluation of the performances with respect to different feature selection methods especially when the ground truth map is unavailable. In this paper, a new method called quantitative multivariate correlation analysis (QMCA) is proposed, which provides a quantitative measure of the useful information in the selected features. QMCA is a combined method of mutual information and correlation information entropy. Using the proposed method, the classification performances of different feature selection methods can be evaluated directly based on the original spectral bands without using the ground truth map. Typical 92AV3C dataset has been applied to the proposed method and the results show that this method is effective, and its conclusions agree with the real classification results in high confidence rate.