Conclusion

This work introduces an analytics system tailored for addressing object counting queries on EO satellites. TargetFuse utilizes both less accurate in space and more accurate ground-based DNN models to determine earth object counts within the constraints of computation and communication. TargetFuse is designed to minimize counting errors under energy and bandwidth constraints. Extensive experiments show that TargetFuse can reduce counting error by 3.4× on average, compared to onboard computing.

本工作提出了一种专用于处理对地观测（EO）卫星上 目标计数查询 的分析系统

该系统名为 TargetFuse，它结合使用精度较低的 星上（in-space） 深度神经网络（DNN）模型与精度较高的 地面 DNN模型，在计算和通信的双重约束下确定地面目标的数量。TargetFuse 旨在能量与带宽约束下，最大限度地减少计数误差

大量实验表明，与纯星上计算（onboard computing）方案相比，TargetFuse 平均可将计数误差降低3.4倍