A methodology for the systems analysis of remote sensing data from natural objects is proposed. This methodology encompasses the solution of interrelated problems of automatic classification in spectral feature space with class aggregation using additional information on the properties of earth’s surface elements, and the synthesis of algorithms for assessing the states of the objects under study. The methodology is based on nonparametric methods and decision-making algorithms, the synthesis of which utilizes kernel probability density estimates. In the first stage of the systems analysis, the initial spectral data characterizing the elements of the study area are broken down into a set of compact observations using nonparametric algorithms for automatically classifying large volumes of statistical information. A class is defined as a compact group of observations of a multivariate random variable, corresponding to a unimodal fragment of its probability density. The detected classes are then combined into groups in the second stage, each with a different distribution pattern for the properties of the earth’s surface elements. For this purpose, an original method for testing hypotheses about the distribution of random variables was developed using a nonparametric pattern recognition algorithm. Based on the information obtained, at the third stage, a training sample is formed for assessing the states of the earth’s surface elements based on their spectral data and a nonparametric pattern recognition algorithm is synthesized. The proposed method allows for modification based on the results of testing the hypotheses under consideration and assessing the states of earth surface elements.
Lapko et al. (2026) studied this question.