neuronelab/s2lcd

Sentinel-2土地覆盖标注数据集是一个专门为遥感图像标注深度学习研究设计的新提出的数据集。它包括1533个图像块，每个图像块大小为224×224像素，来源于Sentinel-2 L2A图像。该数据集确保了温带地区土地覆盖和土地利用类型的多样性，包括森林、山脉、农业用地和城市区域，每个类型都受到不同程度的人类影响。每个图像块都附有五个以COCO格式导出的描述，这些描述使用了丰富的词汇，结合了自然语言和EAGLE词汇表。

The Sentinel-2 Land-cover Captioning Dataset is a newly proposed dataset specifically designed for deep learning research on remote sensing image captioning. It consists of 1533 image patches, each of size 224×224 pixels, derived from Sentinel-2 L2A images. The dataset ensures a diverse representation of land cover and land use types in temperate regions, including forests, mountains, agricultural lands, and urban areas, each with varying degrees of human influence. Each image patch is accompanied by five captions in COCO format, using a vocabulary that combines natural language and the EAGLE lexicon.