A novel approach for spectral-spatial classification of hyperspectral data based on SVM-MRF method

A novel method for spectral-spatial classification of hyperspectral data is proposed. First, a probabilistic pixelwise classification approach is performed using support vector machine (SVM) classifier. Then, erosion technique is used for extracting certain and uncertain pixels from initial classification map. Finally, in order to incorporate spatial information, Markov random field (MRF) regularization process is applied only on uncertain pixels. This concept of using MRF model reduces processing time while improving classification accuracy acceptably. Experimental results are presented for an agricultural hyperspectral data and compared with spectral pixelwise classification and also the conventional SVM-MRF spectral-spatial classification method. The proposed approach is shown better performance when compared to other classification approaches.