跳转至

Related Work

Satellites Scheduling: Many works [27, 38, 39, 49, 62, 63] consider scheduling satellites to point their sensors and maximize the number or value of targets observed. However, these works assume that the target locations are known beforehand and focus on offline scheduling. AB&B [27] considers online scheduling but has several limitations as mentioned in §2.3. Other work [24, 25, 57] focus on a simplified problem, excluding the modeling of actuation time between targets due to their assumption that the sensor is a radar with instantaneous electronic movement capabilities.

卫星调度

许多工作 [27, 38, 39, 49, 62, 63] 研究了如何调度卫星以指向其传感器,从而最大化观测目标的数量或价值。然而, 这些工作假设目标位置是预先已知的,并专注于离线调度 (offline scheduling) 。AB&B [27] 考虑了在线调度,但如2.3节所述,它存在若干局限性。其他工作 [24, 25, 57] 则专注于一个简化的问题,它们假设传感器是具有瞬时电子扫描能力的雷达,因此在其模型中排除了目标之间的驱动时间 (actuation time)。

Heterogeneous Satellite Constellations: Existing leaderfollower designs [7, 18, 58] require offline scheduling and cannot be applied to moving targets, as mentioned in §2.3. Another work [56] aims at a different problem: decreasing the image capture latency for rapid response to disasters.

This work uses two constellations in two different orbit planes: one for imaging and the other as a communication relay between the imaging satellites and Earth.

异构卫星星座

现有的主从式设计 [7, 18, 58] 需要离线调度,并且如2.3节所述,不能应用于移动目标。另一项工作 [56] 则旨在解决一个不同的问题:为实现对灾害的快速响应而降低图像捕获的延迟。该工作使用了位于两个不同轨道平面上的两个星座:一个用于成像,另一个作为成像卫星与地球之间的通信中继。

Superresolution. Superresolution techniques [33, 44] offer a promising means of enhancing image resolution through statistical refinement and data synthesis. However, applying them directly to low-resolution satellite images may not align with our objectives, as it could introduce misleading artifacts for analysts requiring high-fidelity data. On the other hand, super-resolution running on leaders (if made cheap enough) may improve target identification accuracy.

超分辨率 (Superresolution)

超分辨率技术 [33, 44] 通过统计精化和数据合成,为增强图像分辨率提供了一种有前景的方法。然而,将它们直接应用于低分辨率卫星图像可能与我们的目标不符,因为它可能会引入误导性伪影 (misleading artifacts),这对于需要高保真数据的分析师来说是不可接受的。另一方面,如果在主导卫星上运行的超分辨率技术成本足够低,则可能有助于提高目标识别的准确性。

Scheduling in other domains: Recent work has studied architectural and system design techniques for robots [22, 48, 51] and autonomous drones [37, 43]. This work is related to EagleEye because we also consider physical and operational constraints in our system design. However, these efforts are distinct in purpose and mechanism because they aim to optimize hardware architecture for better performance and energy efficiency. Also, satellites differ from robots and drones because they have less freedom of movement: a satellite trajectory is fixed (along the orbit) after launching, and the only thing that can be changed is the satellite orientation.

其他领域的调度

近期的工作研究了机器人 [22, 48, 51] 和自主无人机 [37, 43] 的架构与系统设计技术。这些工作与EagleEye相关,因为我们在系统设计中同样考虑了物理和操作约束。然而,这些工作的目的和机制与我们不同,因为它们旨在优化硬件架构以获得更好的性能和能效。此外,卫星与机器人和无人机不同,因为它们的运动自由度 (freedom of movement) 更少:卫星一旦发射,其轨道轨迹就是固定的,唯一可以改变的是卫星的姿态 (orientation)。

Other aspects of satellite research: Kodan [28] solves the in-orbit computation bottleneck for image processing by training specialized ML models for different geospatial contexts. L2D2 [60] reduces satellite-ground communication latency by adding more low-cost commodity hardware-based ground stations to increase the density of ground stations.

卫星研究的其他方面

  • Kodan [28] 通过为不同的地理空间背景训练专门的ML模型,解决了图像处理的在轨计算瓶颈
  • L2D2 [60] 通过增加更多基于低成本商用硬件的地面站来提高地面站的密度,从而减少了星地通信延迟