DeepSatData

DeepSatData是由帝国理工学院开发的用于训练机器学习模型的大型卫星图像数据集。该数据集利用欧洲空间局的Sentinel2数据，支持大规模数据集的生成，特别适用于深度神经网络的密集分类任务，如语义分割。数据集的创建涉及从Sentinel2数据中自动下载特定区域和时间段的卫星图像，并结合地面实况数据进行全面标注。DeepSatData的应用领域主要集中在地球观测任务，旨在通过机器学习模型解决与地球表面相关的复杂问题。

DeepSatData is a large-scale satellite image dataset developed by Imperial College London for training machine learning models. This dataset leverages Sentinel-2 data from the European Space Agency (ESA) to support the creation of large-scale datasets, and is particularly suitable for dense classification tasks of deep neural networks such as semantic segmentation. The construction of DeepSatData involves automatically downloading satellite images of specific regions and time periods from Sentinel-2 data, and performing comprehensive annotations in combination with ground truth data. The main application fields of DeepSatData are Earth observation missions, with the objective of resolving complex Earth surface-related problems through machine learning models.