SAT-6

SAT-6由总共405,000个图像块组成，每个图像块的大小为28x28，涵盖6个土地覆盖类别-贫瘠的土地，树木，草原，道路，建筑物和水体。选择324,000图像 (包括总数据集的5分之4) 作为训练数据集，选择81,000 (5分之1) 作为测试数据集。与SAT-4类似，训练和测试集是从不相交的NAIP图块中选择的。一旦生成，数据集中的图像将以与SAT-4相同的方式进行随机化。SAT-4和SAT-6的各种土地覆盖类别的规范均采用了国家土地覆盖数据 (NLCD) 算法中使用的规范。

SAT-6 consists of a total of 405,000 image patches, each with a size of 28×28, covering 6 land cover categories: barren land, trees, grassland, roads, buildings, and water bodies. A total of 324,000 images, accounting for four fifths of the entire dataset, are selected as the training dataset, while 81,000 images, accounting for one fifth of the total, are chosen as the test dataset. Similar to SAT-4, the training and test sets are selected from non-overlapping NAIP image patches. Once generated, the images in the dataset are randomized in the same manner as those in SAT-4. The specifications for various land cover categories of both SAT-4 and SAT-6 adopt the standards used in the National Land Cover Database (NLCD) algorithm.