申佳源,成培瑞,王智睿,梁伟,孙显,吴一戎.人工智能赋能遥感：挑战、范式与发展布局[J].中国科学院院刊,2026,41(6):1182-1192.

人工智能赋能遥感：挑战、范式与发展布局

Artificial intelligence empowering remote sensing: Challenges, paradigms and strategic layout

申佳源^1,2
中国科学院空天信息创新研究院 北京 100094;目标认知与应用技术国家级重点实验室 北京 100190
SHEN Jiayuan^1,2
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;National Key Laboratory of Target Cognition and Application Technology, Beijing 100190, China
成培瑞^1,2
中国科学院空天信息创新研究院 北京 100094;目标认知与应用技术国家级重点实验室 北京 100190
CHENG Peirui^1,2
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;National Key Laboratory of Target Cognition and Application Technology, Beijing 100190, China
王智睿^1,2*
中国科学院空天信息创新研究院 北京 100094;目标认知与应用技术国家级重点实验室 北京 100190
WANG Zhirui^1,2*
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;National Key Laboratory of Target Cognition and Application Technology, Beijing 100190, China
梁伟^1,3
中国科学院空天信息创新研究院 北京 100094;中国科学院大学 电子电气与通信工程学院 北京 100049
LIANG Wei^1,3
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
孙显^1,2,3
中国科学院空天信息创新研究院 北京 100094;目标认知与应用技术国家级重点实验室 北京 100190;中国科学院大学 电子电气与通信工程学院 北京 100049
SUN Xian^1,2,3
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;National Key Laboratory of Target Cognition and Application Technology, Beijing 100190, China;School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
吴一戎^1,3*
中国科学院空天信息创新研究院 北京 100094;中国科学院大学 电子电气与通信工程学院 北京 100049
WU Yirong^1,3*
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China

         人工智能;遥感;遥感大模型;物理引导深度学习;智能解译;在轨处理;对地观测

        artificial intelligence;remote sensing;remote sensing foundation model;physics-informed deep learning;intelligent interpretation;onboard processing;Earth observation

        遥感科学与技术作为对地观测与全球变化研究的关键学科，正面临海量多源数据处理、复杂信息精准提取及高时效应用服务的系统性挑战。人工智能技术的快速发展，为破解遥感数据丰富但有效信息挖掘不足的深层困境提供了新的契机。文章坚持问题导向，首先从数据理解、技术方法、科学机理与应用生态4个层面，系统剖析当前遥感学科发展面临的核心挑战；进而梳理人工智能赋能遥感的技术演进脉络，重点阐述自监督与弱监督学习、多模态/跨模态融合、物理引导深度学习、遥感大模型与世界模型四大前沿范式，并以智慧农业、灾害应急、生态安全为典型场景，深入分析其创新价值与实现路径；最后，从基础研究与核心技术攻关2个维度提出系统性发展布局，涵盖多模态全息感知理论、“透视地球”多圈层智能认知、生物启发式遥感智能模型三大基础科学方向，以及遥感标准样本库建设、多源协同观测、全链路自主智能以及星载在轨计算四大核心技术方向。在此基础上，文章研判全球“人工智能＋遥感”的竞争格局与我国所处位置，提出在坚持自主可控的同时应加强开源生态参与和遥感能力的基础设施化。通过构建“智能感知—认知理解—决策支持”的技术链条，推动遥感学科从传统观测分析向智能化、体系化方向演进，为提升我国对地观测科技创新能力与服务国家战略需求提供重要支撑。

        University of Chinese Academy of Sciences, Beijing 100049, China） Remote sensing science and technology, as a key discipline for Earth observation and global change research, faces systemic challenges in processing massive multi-source data, accurately extracting complex information, and delivering high-timeliness application services. The rapid advances in artificial intelligence (AI) provide a new opportunity to address the deep-seated dilemma in remote sensing of being “data-rich but insufficient in effective information mining”. Guided by a problem-oriented approach, this study first systematically analyzes the core challenges facing the development of remote sensing across four dimensions: data understanding, technical methods, scientific mechanisms, and application ecosystems. It then reviews the technical evolution of AI-empowered remote sensing, with a focus on four frontier paradigms — self-supervised and weakly supervised learning, multi-modal and cross-modal fusion, physics-informed deep learning, and remote sensing foundation models and world models, and conducts an in-depth analysis of their innovative value and implementation pathways in three representative scenarios: smart agriculture, disaster emergency response, and ecological security. Finally, a systematic strategic layout is proposed from two dimensions of fundamental research and core technology breakthroughs, covering three fundamental scientific directions, such as multi-modal holographic perception theory, multi-sphere intelligent cognition, and bio-inspired remote sensing intelligent models, as well as four core technology directions: the construction of standardized remote sensing sample libraries, multi-source collaborative observation, full-chain autonomous intelligence, and onboard spaceborne computing. Building on these, the study assesses the global competitive landscape of “AI+remote sensing” and China’s position, arguing that while maintaining independent and controllable development, China should strengthen its engagement in open-source ecosystems and the infrastructuralization of remote sensing capabilities. By building a technology chain of “intelligent perception - cognitive understanding - decision support”, this study advances remote sensing from traditional observational analysis toward an intelligent and systematic direction, providing important support for enhancing China’s earth observation science and technology innovation capability and serving national strategic needs.