Abstract: To meet the needs of future remote sensing satellites for strong autonomy and too many on-orbit collaboration missions. The mechanisms caused by on-orbit autonomous decision-making and intelligent model capability are analyzed, then flow mechanism of a satellite-ground coordination on-orbit decision-making process is established and an on-board decision-making learning method for satellite mission is designed. The learning return function is reinforced by designing the decision-making mission of satellites. The payoff value in reinforcement learning is clarified. The intelligent mission requirement decision-making process of satellites based on a deep Qnetwork-based approach is designed and implemented, a decision-making list is obtained for the autonomous decision-making of satellites. The experiment results show that under the circumstance of 12 kinds of rules, the accuracy of mission attribute configuration reaches over 80%. The reinforcement learning is applied in the mission planning and resource allocation so as to realize such comprehension utilization capabilities as coordination observation, mission coordination, information guidance and others of remote sensing satellites. This research has important theory value for further expanding reinforcement learning theory and the application field of the method.