A Generalizable and Accessible Approach to Machine Learning with Global Satellite Imagery

摘要: Combining satellite imagery with machine learning (SIML) has the potential to address global challenges by remotely estimating socioeconomic and environmental conditions in data-poor regions, yet resource requirements of SIML limit its accessibility use. We show that a single encoding can generalize across diverse prediction tasks (e.g. forest cover, house price, road length). Our method achieves accuracy competitive deep neural networks at orders magnitude lower computational cost, scales globally, delivers label super-resolution predictions, facilitates characterizations uncertainty. Since image encodings are shared tasks, they be centrally computed distributed unlimited researchers, who need only fit linear regression their own ground truth data order achieve state-of-the-art performance.