Towards Autonomous Aircraft Surveillance from Nanosatellites through On-Board Inference and Generative Data Augmentation
A CubeSat-side detector reportedly raises aircraft detection accuracy while staying within edge-hardware limits.
The paper proposes running inference aboard a 6U CubeSat instead of sending raw satellite imagery down for ground processing. To address scarce, imbalanced aircraft datasets, it uses a LoRA-tuned diffusion model to generate minority-class images, then pseudo-labels and merges them with augmented samples. The balanced dataset lifts mean average precision from 77.9% to 82.2%, with the minority-class F1 score rising from 0.683 to 0.811. The quantized detector is reported to fit on-chip memory and project 25-30 frames per second in orbit. ArXiv · AI/CL/LG's note
The paper proposes running inference aboard a 6U CubeSat instead of sending raw satellite imagery down for ground processing. To address scarce, imbalanced aircraft datasets, it uses a LoRA-tuned diffusion model to generate minority-class images, then pseudo-labels and merges them with augmented samples. The balanced dataset lifts mean average precision from 77.9% to 82.2%, with the minority-class F1 score rising from 0.683 to 0.811. The quantized detector is reported to fit on-chip memory and project 25-30 frames per second in orbit. ArXiv · AI/CL/LG's note
score 4