基于Q学习的合成孔径活塞误差校正方法

Piston Error Correction of Synthetic Aperture Systems with Q-Learning

合成孔径成像系统是下一代光学望远镜的发展趋势。当子孔径间的活塞误差校正在一定范围时,能保证系统的成像质量。提出了一种基于Q学习的光学合成孔径系统的校正范围在±λ/2内的精共相活塞误差校正方法。把合成孔径成像系统的图像清晰度作为奖励函数,将子孔径的光程补偿控制动作和状态相结合,在不断学习中提高智能体的决策能力,以达到生成最优控制策略的目的,从而快速校正活塞。通过仿真搭建环境,所建立的智能体模型能在存在活塞误差的环境中获得Q学习中的状态-价值表,并且能通过Q表获取从初始状态到最佳成像性能的快速校正路径序列,同时不受目标场景的限制,验证了该方法的有效性与可靠性。

Synthetic aperture imaging system is the trend of the next generation optical telescope. The imaging quality of the system can be guaranteed only when the piston errors between sub apertures are corrected within a certain range. A precise common phase piston error correction method based on Q-Learning with a correction range of ±λ/2 is proposed. The image clarity of the synthetic aperture imaging system is taken as the reward function,and the optical path compensation control action of the sub aperture is combined with the state to improve the decision-making ability of the agent in continuous learning,so as to generate the optimal control strategy,and quickly correct the piston. Through the simulation environment,the established agent model can obtain the state-value table in Q-learning in the environment with piston error,and can obtain the fast correction path sequence from the initial state to the best imaging performance through the qtable. At the same time,it is not limited by the target scene,which verifies the effectiveness and reliability of this method.