Recognition of arable lands on the territory of Samara region using satellite images for solving land use problems

The paper presents a technology for recognizing arable land from remote sensing images to solve land use problems at the regional level of the Russian Federation. The application of modern deep learning methods to identify the arable land boundaries from both single and a series of medium-resolution Sentinel-2 images is being investigated. According to research, the best quality can be achieved using the UPerNet architecture when extracting multiscale features using Swin Transformer v2 algorithm. The resulting vector layer of arable land is used to solve the problem of detecting illegal plowing of specially protected natural areas. The work makes a significant contribution to improving the efficiency of regional natural resource management systems, demonstrating how the use of artificial intelligence and remote sensing images helps to automate the solution of land use problems.