Development of the Index Paradigm in Remote Sensing of Soil Cover

The aim of the work is the systematic analysis and generalization of the conventional index paradigm of using Earth remote sensing data to assess the state of the soil and vegetation cover. It has been established that the scalar form and the lack of a mathematical basis do not allow the use of conventional vegetation and the similar indices for evaluating the vectors of quantitative indicators of the soil and vegetation cover. At the same time, for making many types of management decisions in agriculture, it is important to construct index images that reflect such qualitative indicators as types of cultivated and weed plants, the presence of plant diseases, damage of crops and soils, and physical and chemical stresses. In terms of informational content, the evaluation of such qualitative states is a procedure for recognizing patterns or classes of soil-and-vegetation complex objects. The subjective empirical approach in choosing the spectral composition of the indices of their combinations, which is currently used, does not currently allow sufficient reliability of such procedures. Therefore, the purpose of the study present is to formalize the process, which enables the empirical approach of constructing indices to be excluded and the entire procedure for their formation for any number and types of recognizable objects to be automated. The basis of formalization is the algorithms for evaluating and selecting the information content of features, followed by the construction of index models, which are linear decision rules for class recognition. The attributes of the classes are the spectral subranges into which the entire spectrum of remote sensing data is divided. The number of informative features is selected from the condition for ensuring the required reliability of recognition of all observed objects (classes).