Phase correlation based hyperspectral image classification using different number of multiple class representatives

In this paper, a phase correlation based supervised classification method for hyperspectral data is proposed. The spectral data of each pixel is initially sub-sampled to increase robustness against noise and spatial variability. Class representatives are extracted using phase correlation based k-means clustering for each class. Phase correlation is used as distance measure in k-means clustering to determine the spectral similarity between each pixel and cluster means. The number of representatives for each class is chosen considering the number of training samples in each class. Classification is performed for each pixel according to the maximum value of phase correlation obtained between samples and the class representatives.