基于深度学习任务去噪的大规模遥感星座 任务规划方法

To address the emerging demand of“massive tasks and efficient computation”in large-scale remote sensing constellations，a novel planning architecture of“task denoising，conflict resolution，and onboard scheduling”is proposed. First，an intelligent task denoising method is designed，introducing the concept of task denoising for the first time in remote sensing task planning. Based on deep learning，non-visible task data are filtered out，thus avoiding meaningless orbital calculations. Second，a cluster-based preferential conflict resolution method is proposed. By designing an onboard redundancy index for conflict resolution，the denoised task data are resolved and uploaded to remote sensing satellites. Finally，a sequence-preferential onboard autonomous task planning method is presented，ensuring a complete onboard planning workflow. Simulation results show that the proposed task denoising method effectively achieves dimensionality reduction of task planning data in large constellations. Compared with traditional planning architectures，it reduces the amount of data to be processed to 21. 4% while preserving complete effective information，and decreases calculation time by 78. 64%. This significantly improves task planning efficiency and meets the novel requirements of large-scale remote sensing constellations for“massive tasks and efficient computation. ”.