Background. Monitoring changes in coastline contours is an actual topic in the field of environmental, geological and information research. However, tasks of this kind are complex and require using modern methods of data processing and analysis, including Earth remote sensing data. One of the modern approaches to solving this class of problems is using machine learning methods, which is the focus of the research in this article. The object of the authors' research is the western coast of the Crimean Peninsula, the study of which by traditional methods has become impossible due to the temporary occupation of the Crimean Peninsula since 2014. In the last decade, the Crimean coastline could have undergone significant changes as a result of anthropogenic activities (including those related to military operations) and landslide-abrasive processes. In this study, the authors limit the study to changes in the coastline of the western part of the Crimean Peninsula over the last decade. Methods. Authors used CNN models (U-Net model) to effectively recognize the coastline and its boundaries in satellite images without the need for manual vectorization. Results. The research involved developing Python code to automatically generate reports including network accuracy (0.95) and loss function (0.19), facilitating the evaluation of different approaches and methods. Additionally, the study created scripts for using the trained network in the task of semantic segmentation and translating the result of the segmentation model into a vectorized result of the coastline contours of the Crimean Peninsula, which was represented as a probability raster. Conclusions. The use of this approach is useful for monitoring changes in the coastline of rivers, seas and lakes throughout Ukraine.