BD-Sat

BD-Sat provides a high-resolution dataset that includes pixel-by-pixel LULC annotations for Dhaka metropolitan city and the rural/urban area surrounding it. With the strict and standard procedure, the ground truth is made using Bing-satellite imagery at a ground spatial distance of 2.22 meters/pixel. Three stages well-defined annotation process has been followed with the support from geographic information system (GIS) experts to ensure the reliability of the annotations. We perform several experiments to establish the benchmark results. Results show that the annotated BD-Sat is sufficient to train large deep-learning models with adequate accuracy with five major LULC classes: forest, farmland, built-up, water, and meadow.