Extended Subspace Method for Remote Sensing Image Classification

This study proposes an extended subspace method (ESM) in feature extraction and dimension-reduction problems for land cover classification of hyperspectral and multi-spectral remote sensing images. The main idea of our method is to use a multiple similarity method (MSM) onto an averaged learning subspace method (ALSM) and makes use of fidelity value criteria in the selection of the optimal subspace dimensions. This method is compared with the support vector machine (SVM) method using Compact Airborne Spectrographic Imager-2 (CASI-2) hyperspectral remote sensing data. Experimental results show that ESM is a valid and effective alternative to other pattern recognition approaches for the classification of remote sensing data.