An improved marker selection method for hyperspectral image segmentation and classification

Abstract

Recently, a new approach for spectral-spatial classification of hyperspectral images has been proposed by Tarabalka et al. This approach is based on the Minimum Spanning Forest (MSF) grown from automatically selected markers by using the Support Vector Machines (SVM) classification. This paper aims at improving this approach by means of a new method for the selection of markers. This method is a combination of SVM and multi-layer perceptron (MLP) neural network classifiers. In the proposed method, the most reliable pixels, i.e. markers, are extracted from the classification maps and used to build the MSF. Three scenarios are evaluated for the first stage of marker selection: SVM, MPL and combination of SVM and MPL. Experimental results on two benchmark hyperspectral datasets demonstrate that the proposed method significantly improves the classification accuracies compared to the approach based on the SVM classification.