A large dataset with CNN-based segmentation analysis for land cover from large river basins

Land cover and segmentation play an important role to ensuring the safety of people's lives and property from floods in large river basins. Large river basins are generally located in rural and urban areas, with different distribution and diversity of feature classes. However, most of the current remote sensing image segmentation algorithms focus on general targets, for large basin specific objects such as dams and river beaches, algorithms trained on general datasets have poor performances. For land cover task in large river basins, a land segmentation dataset has been proposed, including eight categories of large river basin at sub-meter spatial resolution. The experimental results show that the dataset can effectively support the training of the large river basin segmentation algorithms.