Evaluating Energy and Visibility Trade-offs in LEO Satellite Edge Computing for Airplane Detection

Abstract

This thesis investigates the efficiency of various Low Earth Orbit (LEO) configurations for autonomous airplane detection using onboard edge computing in space. The study evaluates the trade-off between maximizing target visibility—defined as daytime passes suitable for optical detection over Incheon International Airport—and minimizing energy consumption for satellite operations. A baseline Sun-Synchronous Orbit (SSO) is compared against multiple Non-Sun-Synchronous Orbits (NSSOs) with varying inclinations (45°–110°), RAANs, and arguments of perigee. For each configuration, orbital visibility is simulated over 1-day and 7-day windows using Python-based tools with J2 perturbation modeling validated against GMAT. Daytime visibility is filtered using local KST-based illumination constraints. To quantify operational efficiency, a composite scoring system is introduced that aggregates normalized visibility metrics, followed by an energy modeling framework that estimates consumption per image strip and per orbit. The model incorporates realistic assumptions: an optical payload based on KOMPSAT-3 specifications, onboard inference with YOLOv5n on NVIDIA Jetson Xavier NX, and S-band downlink of detection results. A simplified solar generation model evaluates the power budget. The findings reveal that mid-inclination NSSOs (especially at 70°) strike the most effective balance between data yield and energy sustainability. The methodology provides a scalable framework for orbit design tailored to edge-AI missions requiring both high revisit rates and energy efficiency.