GFM-Bench/BigEarthNet

BigEarthNet是一个大规模的多标签分类基准数据集，由Sentinel-1（雷达）和Sentinel-2（光学）卫星图像组成。数据集经过预处理，将所有Sentinel-2通道上采样到120x120像素并连接在一起。原始的43个土地覆盖类别被映射到19个更广泛的类别。数据集分为训练集、验证集和测试集，包含269,695个训练样本、123,723个验证样本和125,866个测试样本。数据集特征包括雷达图像、光学图像、分类标签、雷达和光学通道的中心波长以及图像的空间分辨率。

BigEarthNet is a large-scale benchmark dataset for multi-label classification, derived from Sentinel-1 (radar) and Sentinel-2 (optical) satellite imagery. The dataset has been pre-processed by upsampling all Sentinel-2 channels to 120x120 pixels and concatenating them together. The original 43 land cover classes are mapped to 19 broader categories. The dataset is split into training, validation, and test sets, containing 269,695 training samples, 123,723 validation samples, and 125,866 test samples. Dataset features include radar image, optical image, classification label, central wavelengths of radar and optical channels, and the spatial resolution of images.