Hyperspectral image segmentation using the Dirichlet mixture models

In this study, segmentation of hyperspectral images which is a multidisciplinary subject was proposed using Dirichlet mixture models. Due to the computational complexity and high volume and dimensional nature of hyperspectral images, principal component analysis (PCA) and its kernelized version kernel PCA (KPCA) were used in dimension reduction stage. Pre-segmentation step was realized with a selected sub-sampled dataset from all data; then segmentation of whole scene is accomplished by support vector machines (SVMs) and k-nearest neighbors (k-NN) methods. Obtained results are evaluated with k-means and fuzzy c-means algorithms by power of spectral discrimination (PWSD) metrics.