A decision fusion framework for high-resolution remote-sensing image classification

Abstract

Classification of high-resolution remote-sensing images is a challenging research area. In this paper we proposed a novel decision fusion framework to combine bag of features (BOF) based classifiers. The proposed framework, can also be used in multi category image classification applications. A single voting algorithm is used for decision fusion and an ambiguity detection module is used to determine the ambiguous situations. An ambiguous situation will occur during multi-category voting, where more than one class got the maximum votes, and also when the number of the same votes doesn't exceeds a desired threshold. To resolve this situation we proposed to use the earth mover's distance (EMD) which is a metric for histogram matching. Indeed, we used the EMD to compare the BOF based histogram of images with the centroid classes. Finally, to evaluate the proposed framework, we used a multi-category remote-sensing image dataset and compared the proposed approach with several other similar approaches with BOF based classifiers. The experimental results demonstrate the effectiveness of the proposed framework.