Recognition of hyperspectral images with use of cluster ensemble and semisupervised learning

We suggest a method for hyperspectral image analysis on the basis of semi-supervised learning. The main idea is to divide the process of training of a classifier into two stages. First of all, with usage of cluster ensemble algorithms, variants of image segmentation are obtained. On their basis, the averaged co-Association matrix is calculated. On the second stage, a classifier is constructed on labeled pixels using similarity based learning algorithms with the given matrix as input. An example of the application of the method for analysis of hyperspectral images is given. It is shown that the suggested algorithm is more robust to noise than the standard support vector machine method.