Improving the Efficiency of Remote Sensing Data Interpretation by Analyzing Neighborhood Descriptors

Abstract

Introduction. In evaluating the space-time structure of the Earth’s surface, the data of remote sensing of the Earth become more important. Increasing the effectiveness of space survey analysis tools is possible through studying the problem of obtaining an integrated space-time characterization of the state of lands. The purpose of this study is to improve the accuracy of the automated analysis of remote sensing data by taking into account the invariant and dynamic descriptors of the vicinity. Materials and Methods. In order to improve the accuracy of the remote sensing data classification, a computation of complex space-time characteristics of the state of the lands was conducted based on the system analysis of data characterizing the dynamic and invariant states of the territory surrounding the geophysical site. The formalization of this process includes methods for calculating a set of numerical descriptors of the neighborhood: local entropy, local range, standard deviation, color moment, histogram of hues, and color cortege. A technique for calculating a complex descriptor based on the Fisher vector is described. To approbate the solution, a plan for the experiment was drawn up and a sample of the initial data was sampled. Results. The approbation of the methodology and the algorithm developed on its basis, implemented as a set of programs, on the test polygon system showed a variation in the classification accuracy in the range of 81–89% (without regard to the neighborhood), and taking into account the neighborhood, it increases to 91–97%. It is revealed that a significant increase in the radius of the analyzed neighborhood leads to a decrease in the classification accuracy. Conclusions. The application of the developed set of programs allows for the rapid implementation of modeling of spatial systems for the purpose of thematic mapping of land use and analyzing the development of emergency situations. The developed methodology for analyzing lands with regard to the descriptors of the neighborhood makes it possible to improve the accuracy of classification.