Preliminary Design and Methodological Framework for an AI-Driven Decision Support System in Earth Observation Satellites

This paper presents a preliminary design and evaluation of an AI driven decision support system for optimizing satellite imaging under cloud interference. Cloud cover renders over eighty percent of captured images unusable, leading to significant operational delays and economic losses. To address this challenge, the proposed system develops a drone-based simulation platform that integrates real time cloud detection, motion tracking, and adaptive mission planning. The system employs a deep learning model for cloud detection and a lightweight tracking algorithm for cloud movement prediction, both running on an embedded GPU unit. An onboard planner analyzes cloud maps to select optimal imaging positions and issues flight commands through a standard autopilot controller. Experimental results demonstrate cloud detection accuracy above ninety percent, an increase in imaging success rate from forty three percent to eighty eight percent, and a substantial reduction in redundant data captures. Low inference latency and efficient resource usage confirm the feasibility of onboard deployment. These findings highlight the potential of embedded AI to enable autonomous satellite operations, reduce dependence on ground control, and improve image quality, laying the groundwork for future fully autonomous Earth observation systems.