Sub-pixel Mapping Based on SVM of Hyperspectral Remotely Sensed Imagery

In this paper, an image sub-pixel mapping algorithm based on support vector machine (SVM)has been presented for hyper spectral imagery. Since the total variation(TV) model is classic Edge-preserving smoothing filter, the authors introduce this model as a presmoothing to improve accuracies of spectral unmixing and sub-pixel mapping. Also, according to the spatial correlation theory, our algorithm not only considers the impact of the abundance for the current pixel on sub-pixel classification, but also takes the effect of adjacent pixels into account. In addition, to improve the efficiency of our algorithm, we propose to decrease the number of samples by eliminating pure pixels during the training and testing procedure in supervised classification. Experimental results on real-world hyper spectral remote sensing dataset show the validity of our algorithm on both visual inspection and quantitative analysis.