Procedure for decomposing the values of two-dimensional spectral features of remote sensing based on the analysis of correlation coefficient components

A procedure was developed for decomposing the values of two-dimensional spectral features according to the components of their correlation coefficients. This procedure, which is similar to automatic classification algorithms, is based on analyzing the proposed indicator-products of normalized spectral features and their probability density. A nonparametric Rosenblatt-Parzen estimator is used to reconstruct probability density from the initial statistical data. The specificity of the proposed indicator and its user-selected threshold values enable the generation of options for the initial statistical data decomposition and mapping of the obtained results during a computational experiment. By means of the human-machine procedure for decomposing the values of two-dimensional spectral features, it is possible to circumvent the problem of solving optimization problems when applying automatic classification algorithms and to use information about the dependence between spectral features in the elements of the Earth's surface. The authors present the results of applying the procedure in processing remotely sensed forest data, as well as their comparison with the initial data. The spectral features that primarily determine the decomposition into dead stands and other forest conditions are established. The obtained results can be used to generate sets of spectral features in assessing the condition of natural objects.