Field Boundary Instance Segmentation–22M (FBIS-22M)

FBIS-22M是由欧洲空间局等机构联合创建的大规模农田边界实例分割数据集，作为该领域迄今最大的基准数据集，包含67.29万张多分辨率卫星影像块（0.25-10米）和2290万个经过验证的农田实例。数据集通过严谨的标注流程整合了多源遥感数据，覆盖从细碎田块到大型农场的多样化农业景观。该数据集主要应用于农业遥感监测领域，旨在解决传统方法在田块边界提取中存在的完整性不足、相邻田块合并及扩展性受限等关键问题，为精准农业管理和粮食安全评估提供数据支撑。

FBIS-22M is a large-scale farmland boundary instance segmentation dataset jointly created by the European Space Agency and other institutions. As the largest benchmark dataset in this field to date, it contains 672,900 multi-resolution satellite image patches with a spatial resolution ranging from 0.25 to 10 meters, as well as 22.9 million verified farmland instances. The dataset integrates multi-source remote sensing data through a rigorous annotation workflow, covering diverse agricultural landscapes spanning from finely fragmented croplands to large-scale farms. This dataset is primarily applied in the field of agricultural remote sensing monitoring, aiming to address key shortcomings of traditional field boundary extraction methods, including insufficient extraction completeness, merging of adjacent croplands, and limited scalability, and provides data support for precision agricultural management and food security assessment.