Development and Comparison of Deep Learning Models for Segmentation of Satellite Images for Localization of Soil Erosion

This work is devoted to solve the problem of automatic determination of the geographic location of soils showing signs of degradation. The work carried out a detailed study of the existing literature on the subject, considered algorithms and approaches to solving the problem and their advantages and disadvantages, as well as proposed evaluation criteria and a basic algorithm for comparison with the results of further research and experiments. This approach provides an adequate assessment of the quality of image segmentation software development. Since machine learning algorithms require pre-processing of data, and there are also certain features of working with satellite images, a number of functions have been developed that can be used by such algorithms in the future. The result of the work is a developed software package for downloading and processing satellite images, as well as training and evaluation of deep learning models.