FEDERATED LEARNING FOR ON-ORBIT EARTH OBSERVATION ANALYTICS IN MILITARY SATELLITE CONSTELLATIONS

Abstract

Earth observation (EO) systems play a central role in modern military intelligence, surveillance, and reconnaissance by providing persistent, wide-area sensing of operational environments. However, conventional EO processing architectures rely heavily on ground-based analysis, leading to increased latency, high communication bandwidth requirements, and reduced resilience in contested or degraded communication scenarios. This paper explores federated learning as a system-level enabler for on-orbit EO analytics in distributed satellite constellations, allowing satellites to perform local inference and limited model adaptation while coordinating globally through the exchange of model updates rather than raw sensor data. We present a reference architecture for federated learning–enabled EO systems, analyze its operational advantages for time-critical military applications, and discuss key cybersecurity and trust challenges arising from distributed learning across space and ground segments. The paper argues that federated on-orbit analytics can significantly enhance responsiveness, robustness, and scalability of military EO systems when integrated within a secure, system-aware design framework.