Advancing Multi-Class Object Detection from LEO/VLEO: Model Evaluation and Onboard Deployment tailored for a 16U CubeSat

On-board AI-based image recognition for Very Low Earth Orbit (VLEO) and Low Earth Orbit (LEO) missions enables the potential for high-resolution Earth Observation while minimizing data transmission needs. This paper presents a multi-class object detection approach using deep learning models. A custom Earth Observation dataset was created, considering orbital and sensor characteristics, to evaluate the model’s performance. This work provides proof of concept for real-time object detection in LEO/VLEO missions including deployment on to the based target edge hardware, addressing the challenges of limited computational resources and the need for automated image analysis. These results highlight the potential for autonomous payload operations for Earth Observation in future VLEO and LEO missions.