The Power of Remote Sensing in Earth Observation

Remote Sensing (RS) stands as a pivotal tool for monitoring and interpreting our planet, playing a vital role in fields ranging from geoscience to economics and humanitarian efforts.

AI Breakthroughs Meets Complex Challenges

While artificial intelligence (AI) and deep learning have dramatically advanced remote sensing capabilities, numerous challenges remain. The complexity of Earth’s environments, diverse sensor technologies, and variations in data quality all complicate the development of smarter RS systems.

Introducing Remote Sensing Foundation Models

Recent developments in large Foundation Models (FMs) showcase AI's capabilities across various fields, boasting impressive generalizability and zero-shot learning. Nevertheless, these models have struggled with non-optical remote sensing data, which hampers their effectiveness. This gap has sparked enthusiasm for creating specialized Remote Sensing Foundation Models (RSFMs) tailored for the multifaceted demands of Earth Observation.

A Comprehensive Survey of RSFMs

This survey meticulously examines the burgeoning field of RSFMs. It starts by outlining the motivations behind their creation and delving into foundational concepts, followed by a systematic categorization and review of current RSFM studies. Key contributions are highlighted across different domains, including Visual Foundation Models (VFMs), Visual-Language Models (VLMs), and Large Language Models (LLMs).

Benchmarking and Future Directions

The survey benchmarks these models against publicly available datasets, identifies prevailing challenges, and lays out promising research avenues for the future. As RSFMs continue to evolve, they hold the potential to revolutionize how we observe and understand our planet.

Leading the Way in AI and Earth Observation

Noteworthy contributors to this important survey include Aoran Xiao, Weihao Xuan, Junjue Wang, Jiaxing Huang, Dacheng Tao, Shijian Lu, and Naoto Yokoya, showcasing a collaborative effort to bridge the gap between AI and Earth observation.