EarthView
收藏资源简介:
EarthView是一个专为遥感数据自监督学习设计的大规模数据集,旨在提升地球监测任务中的深度学习应用。该数据集由Satellogic、NEON和Sentinel等多个来源的遥感图像组成,总像素数达到15万亿,涵盖了从2017年到2022年的五年数据。数据集包含多种分辨率的图像数据,涵盖了高光谱、多光谱、地形数据、分割图和时间结构等多种模态。数据集通过HuggingFace平台以parquet格式提供,便于研究人员使用。该数据集的应用领域包括气候变化监测、自然灾害预警、农业和城市规划等,旨在通过自监督学习提升地球监测算法的性能。
EarthView is a large-scale dataset specifically designed for unsupervised learning of remote sensing data, aiming to enhance the application of deep learning in Earth observation tasks. Composed of remote sensing images from multiple sources including Satellogic, NEON, and Sentinel, the dataset boasts a total of 1.5 trillion pixels and spans five years of data from 2017 to 2022. It encompasses a variety of image data resolutions and includes high-spectral, multispectral, topographic data, segmentation maps, and temporal structures across multiple modalities. The dataset is provided in parquet format via the HuggingFace platform, facilitating ease of use for researchers. The application domains of this dataset include climate change monitoring, natural disaster early warning, agriculture, and urban planning, with the goal of improving the performance of Earth observation algorithms through unsupervised learning.

- 1EarthView: A Large Scale Remote Sensing Dataset for Self-Supervision计算机视觉中心, Satellogic, Mila, 蒙特利尔大学, ServiceNow研究, 苹果研究 · 2025年



