A Framework of Multiple Kernel Ensemble Learning for Hyperspectral Classification

Hyperspectral image classification has been a very active area of research in recent years. Multiple kernel learning (MKL), ensemble learning are promising family of machine learning algorithms, have been applied extensively in hyperspectral image classification. However, many MKL methods often formulate the problem as an optimization task. Due to the high computational cost of solving the complicated optimization problem, improve the efficiency of MKL, in this paper, an ensemble learning framework, SMKB (Stochastic Multiple Kernel Boosting), which applies Adaptive Boosting (AdaBoost), stochastic approach to learning multiple kernel-based classifier for multi-class classification problem, is presented. We examine empirical performance of proposed approach on benchmark hyperspectral classification data set in comparison with various state-of-the-art algorithms. Experimental results show that SMKB is more effective, efficient than traditional MKL techniques.