Textural Feature Extraction and Ensemble of Extreme Learning Machines for Hyperspectral Image Classification

Abstract

The use of textural information is very important in classification of hyperspectral images. In this paper, we used local binary patterns, histograms of directional gradients and Gabor filters for extract the textural properties of the hyperspectral images. Then, we have proposed a two-level feature combination method on the obtained textural properties. It is aimed to increase the classification results on hyperspectral images with using radial based extreme learning machine on the fused features. On this purpose, it has also been proposed to combine decisions made by extreme learning machines. These methods have been applied on Indian Pine hyperspectral images with ground truth information and it is observed that they obtain more robust results than traditional alternative methods.