Multi-LEO Satellites Cooperative Electromagnetic Sensing via Deep Learning: Vision and Challenges

The electromagnetic (EM) environment is becoming increasingly complex, and competition for spectrum resources continues to intensify. Traditional ground-based EM monitoring systems, constrained by limited spatial coverage, are insufficient to meet the growing demand for wide-area EM sensing. In contrast, space-based EM sensing provides inherently broader observation coverage. Low Earth orbit (LEO) satellites offer reduced propagation latency relative to medium-Earth-orbit (MEO) and geostationary-orbit (GEO) satellites, flexible deployment, and high spatiotemporal resolution, making them particularly suitable for constructing a global space-based EM sensing network. Ultra-dense LEO constellations can collaboratively receive terrestrial EM signals and share data across nodes, improving sensing continuity, robustness, and coverage. However, the massive, heterogeneous, and dynamic EM data generated by distributed satellites challenges traditional signal processing, particularly in cross-node fusion and spatiotemporal feature extraction. Deep learning (DL), with its end-to-end modeling and feature learning capabilities, shows strong potential in modulation recognition, interference identification, and passive localization. Integrating DL with multi-LEO cooperation paves the way for accurate, scalable, and intelligent EM sensing. This article proposes an intelligent architecture for multi-LEO satellite cooperative EM sensing, analyzes key challenges in applying DL, and explores potential solutions such as physics-informed priors, self-supervised learning, and multi-task collaboration, while outlining future research directions.