Stitching Satellites to the Edge: Pervasive and Efficient Federated LEO Satellite Learning

TL;DR

This work tackles the challenge of bringing privacy-preserving AI to LEO satellite constellations by integrating federated learning with satellite edge computing. It introduces personalized divide-and-conquer learning to convert multi-class tasks into lightweight binary problems and orbital retraining to build and refine orbital models before PS transmission, drastically cutting communication rounds. Experiments on on-board Jetson Nano hardware using EuroSat, MNIST, CIFAR-10, and CIFAR-100 demonstrate rapid convergence (around 2–4.6 hours) with high accuracy (up to ~96%) and low energy consumption (as low as ~1.38 W per satellite), outperforming multiple baselines. The approach offers practical scalability for resource-constrained space-edge environments and paves the way for pervasive, efficient space AI with secure, on-orbit training and aggregation.

Abstract

In the ambitious realm of space AI, the integration of federated learning (FL) with low Earth orbit (LEO) satellite constellations holds immense promise. However, many challenges persist in terms of feasibility, learning efficiency, and convergence. These hurdles stem from the bottleneck in communication, characterized by sporadic and irregular connectivity between LEO satellites and ground stations, coupled with the limited computation capability of satellite edge computing (SEC). This paper proposes a novel FL-SEC framework that empowers LEO satellites to execute large-scale machine learning (ML) tasks onboard efficiently. Its key components include i) personalized learning via divide-and-conquer, which identifies and eliminates redundant satellite images and converts complex multi-class classification problems to simple binary classification, enabling rapid and energy-efficient training of lightweight ML models suitable for IoT/edge devices on satellites; ii) orbital model retraining, which generates an aggregated "orbital model" per orbit and retrains it before sending to the ground station, significantly reducing the required communication rounds. We conducted experiments using Jetson Nano, an edge device closely mimicking the limited compute on LEO satellites, and a real satellite dataset. The results underscore the effectiveness of our approach, highlighting SEC's ability to run lightweight ML models on real and high-resolution satellite imagery. Our approach dramatically reduces FL convergence time by nearly 30 times, and satellite energy consumption down to as low as 1.38 watts, all while maintaining an exceptional accuracy of up to 96%.

Figures (6)

• Figure 1: Personalized Learning via Divide-and-Conquer.
• Figure 2: Illustration of Orbital Retraining (Section \ref{['Sec:SinkSat']}).
• Figure 3: Our implementation, training, and testing on Jetson Nano with experimental setup of essential peripherals.
• Figure 4: Orbital model accuracy comparison across global communication rounds and orbital epochs. The upper subplot shows the accuracy of orbital models over two global communication rounds. The middle subplot illustrates the evolution of orbital model accuracy for orbit #1 in global round #1 over 5 orbital epochs. The lower subplot shows the training accuracy for each satellite model on orbit #1 in orbital epoch #1 and global round #1. The high accuracy is attributed to our DnC approach that converts complex tasks into binary tasks.
• Figure 5: Confusion matrix that compares 10 predicted and ground-truth classes for 5400 test images.