Метод Селекции Объектов На Гиперспектральном Изображении На Основе Анализа Их Контуров

Abstract

A new method of spectral selection of given objects on hyperspectral images is considered. At the first stage of the method, hypotheses are tested using the Neyman-Pearson criterion about the presence of object contours in neighboring pixels relative to the simple alternative of their absence consistently over all spectral components. If a decision is made about the presence of a contour in at least one spectral channel, these pixels are analyzed at the second stage with respect to their distribution over the spectral range according to the criterion of maximum a posteriori probability density. Given the values of the mathematical expectation of the gradient difference between the spectral components, hypotheses are formed about the presence or absence of the contour of the desired object. The decision is made on the basis of a comparison of the decision statistics with the likelihood functions. The characteristics of detection and the results of experiments performed on real images are presented.