[论文解读] Artificial intelligence to advance Earth observation: : A review of models, recent trends, and pathways forward

Devis Tuia, Konrad Schindler+12 more|arXiv (Cornell University)|May 15, 2023

Computational Physics and Python Applications被引用 38

一句话总结

一篇观点性综述，评估 AI/ML、计算机视觉与先进处理如何改变对地观测，强调混合物理-ML、基于知识的 AI，以及伦理/以用户为中心的考量。

ABSTRACT

Earth observation (EO) is a prime instrument for monitoring land and ocean processes, studying the dynamics at work, and taking the pulse of our planet. This article gives a bird's eye view of the essential scientific tools and approaches informing and supporting the transition from raw EO data to usable EO-based information. The promises, as well as the current challenges of these developments, are highlighted under dedicated sections. Specifically, we cover the impact of (i) Computer vision; (ii) Machine learning; (iii) Advanced processing and computing; (iv) Knowledge-based AI; (v) Explainable AI and causal inference; (vi) Physics-aware models; (vii) User-centric approaches; and (viii) the much-needed discussion of ethical and societal issues related to the massive use of ML technologies in EO.

研究动机与目标

• Explain how AI methods are transforming Earth observation across modelling, understanding, and communication.
• Discuss challenges unique to EO data (data fusion, ground truth sparsity, noisy labels, scale) and potential AI solutions.
• Highlight system architectures, interoperability, and ethical/trust considerations for AI in EO.

提出的方法

• Survey of computer vision and ML techniques applied to EO tasks such as translation, segmentation, and retrieval in multi-sensor contexts.
• Discussion of learning paradigms for EO, including supervised, self-supervised, semi-supervised, and active learning.
• Review of AutoML, NAS, and domain/adaptation methods tailored to spatio-temporal EO data.
• Examination of advanced processing architectures (EO ecosystems, federated learning, interoperability, scalable processing) to enable large-scale EO AI.
• Integration of knowledge-based AI, causal inference, and physics-aware ML to improve interpretability and reliability.
• Consideration of user-centric design, trust, ethics, and data governance in AI-for-EO workflows.

Figure 1: Conceptual overview of this perspective paper: (a) different levels of algorithms emerge from the areas of machine learning (ML) and interact with computer vision (CV), computer science, and statistics to learn patterns and associations from observational data. The models must integrate do

实验结果

研究问题

• RQ1 What ML and CV approaches are most effective for processing and understanding EO data across sensors and resolutions?
• RQ2 How can hybrid physics-based and data-driven models improve EO tasks and reduce data requirements?
• RQ3 What architectures and learning paradigms best support scalable, interoperable, and trustworthy AI in EO?
• RQ4 How can AutoML, domain adaptation, and continual learning address data sparsity, noise, and distribution shifts in EO?
• RQ5 What ethical and user-centered considerations must guide AI adoption in Earth observation?

主要发现

• EO benefits from data fusion and representation learning to integrate diverse sensor data.
• Hybrid modelling combining physical knowledge with ML can improve robustness and data efficiency in EO.
• AutoML and NAS hold promise for finding task- and data-specific models in EO, especially for spatio-temporal data.
• Domain adaptation and continual learning are essential to handle distribution shifts and growing EO archives; open-world and memory-based strategies are relevant.
• Federated learning offers privacy-preserving, distributed training but faces non-iid data challenges typical in EO.