Abstract:Aiming at the problem of multiple satellites collaborative planning with unobservable ground moving targets, this paper studies the online imaging scheduling method based on improved deep reinforcement learning algorithms. First, based on the satellite imaging coverage calculation method, an imaging strip partitioning algorithm is proposed. Second, based on the visible time window, strip and satellite orbit information, a multiple satellites cooperative observation mission planning model based on the partially observable Markov decision process(POMDP) is established. Then, a proximal policy optimization algorithm with an action mask and advantage normalization mechanism is proposed, which improves the convergence rate of the algorithm for solving the partial strip coverage task area scheduling problem. Finally, the correctness and superiority of the proposed algorithm are verified by three sets of simulations.