Mapping of land use structure based on deciphering of multi-zone space images for the purposes of geoecological zoning of cultural landscape

The article reveals the issues of application of land use structure mapping techniques using algorithms of automated interpretation of Earth remote sensing data (multispectral space images) for the purposes of regional geoecological zoning. Research conducted in 2024. Landsat satellite images were used as Earth remote sensing data. Scientific research was conducted using neural network methods based on test sites on the territory of the Republic of Mordovia, which have different land use regimes. The test sites have different land use regimes. The experiments were conducted in the ScanEx Image Processor software package. The most effective topology of the direct distribution neural network (input channels, number of neurons in the hidden layer) was determined. Error matrix calculations were performed on control plots to assess the accuracy of interpretation. Neural networks with the proposed topologies showed high decoding accuracy. Land Cover land use class mapping was performed for a test site with an intensive land use pattern. The average proportion of pixels not assigned to classes was 94.8 %. The overall classification accuracy with pixels not assigned to classes was 0.4 %. Mapping of natural geosystems was carried out for the territory of the polygon with a nature conservation regime and weak economic development. The accuracy rates were 97.0% and 1.3%, respectively. The presented approaches to land use mapping and the results of other research made it possible to develop a model of geoecological zoning of the cultural landscape of the region with the identification of zones of ecological balance and an economic framework. The results can be used to compile and modify cartographic material in territorial planning documents at the regional and municipal levels.