Classification of Hyperspectral Images with Multiple Kernel Extreme Learning Machine

Abstract

In this work, it is aimed to increase the classification success of hyperspectral images with using multiple kernel extreme learning machine (MK-ELM) by obtaining optimal convex combination of predefined kernel functions. The use of intuitive iterations instead of complex optimization processes and the facility of multi-class classifications make MK-ELM more advantageous than support vector machine (SVM) based multiple kernel learning (MKL) methods. MK-ELM applied to Pavia University hyperspectral scene that has ground truth information with using 11 different Gaussian and polynomial kernels constructed with various parameters and than obtained results are presented comparatively along with the state-of-the-art SVM based MKL methods.