Nonparametric algorithm for automatic classification of remote sensing data

A nonparametric algorithm for automatic classification of large-volume statistical data is proposed. The algorithm under consideration assumes compression of initial information based on decomposition of multidimensional feature space. As a result, a large statistical sample is transformed into a data array composed of the centers of multidimensional sampling intervals and their corresponding frequencies of random variables. The information obtained is used in the synthesis of the regression estimate of the probability density. A class is understood as a compact group of observations of a random variable corresponding to a unimodal fragment of the probability density function. On this basis, a nonparametric automatic classification algorithm is developed which is based on the sequential procedure for checking the proximity of the centers of multidimensional sampling intervals and the ratios between the frequencies of belonging of random variables from the original sample to these intervals. To improve the computational efficiency of the proposed automatic classification algorithm, a multithreaded method of its software implementation is used. The practical significance of the developed algorithm for automatic classification is confirmed by the results of its application for assessing the state of the forests areas using remote sensing data.