The recognition of borders in satellite data is critical for many remote sensing applications, including land cover categorization, urban planning, and environmental monitoring. We present the RSUNet-A architecture, a unique strategy that combines the U-Net[1] design with recursive skip connections, in order to overcome this. The RSUNet-A model delivers extremely precise border recognition in satellite data by combining both local and global contextual information. Our tests show that the RSUNet-A model beats previous techniques in terms of localization and border correctness, constituting a substantial development in this area. By improving border recognition in satellite images, the suggested architecture has the potential to improve geospatial analysis and decision-making procedures