Thematic classification of hyperspectral images using conjugacy indicator

We consider an algorithm of hyperspectral images thematic classification using conjugacy indicator as a proximity measure. This measure is a generalized spectral angle mapper (SAM) implemented in hyperspectral imagery processing software ENVI. In this case, we use the cosine of an angle between considered vector and subspace, which is spanned by class vectors, instead of the cosine of an angle between considered vector and the mean vector of the class. Paper describes modification of a method based on partitioning of the class into subclasses and based on reduction of vectors to zero mean value. The results of synthetic experiments show higher classification quality than SAM.