Remote Sensing Image Segmentation

Remote sensing image segmentation assigns classes or object boundaries to pixels in satellite and aerial images, supporting tasks such as land-cover mapping, flood and wildfire monitoring, and building or road extraction. Common approaches include U-Net-style convolutional networks, transformers and state-space models, and fusion of optical, SAR, elevation, or time-series data; newer work also adapts geospatial foundation and vision-language models. Benchmarks such as ISPRS Vaihingen and Potsdam, LoveDA, OpenEarthMap, and Sen1Floods11 test accuracy across varied sensors, regions, and applications. A central challenge is reliable transfer under geographic, seasonal, and sensor shifts, while preserving fine boundaries and balancing annotation needs, computational cost, and performance on rare classes.