| According to Rayleigh criterion,the resolution of the telescope increases with the aperture of the system when the wavelength of light passing through the optical system is constant.However,with the restriction of the huge costs brought by the increasing system aperture,the manufacturing technology,system installation as well as the payload,it is difficult for the traditional single-aperture telescope to continue its development and breakthrough.In this context,synthetic aperture imaging technology emerges,which is capable of obtaining the equivalent imaging capability of large aperture optical system through a certain arrangement of multiple separated mirrors.However,the consequent co-phasing error will be one of the key factors that hinder the high-resolution imaging performance of the system.Therefore,the detection and correction of co-phasing error is of great practical significance.In this paper,the piston error in the co-phasing error is studied by using reinforcement learning framework and the simulation results are presented,which provides an effective solution to the co-phasing problem of synthetic aperture.Firstly,the research progress of the existing optical synthetic aperture imaging system has been reviewed,and the current detection methods of co-phasing error and their limitations are summarized.According to the main problems and challenges of the current co-phasing detection methods,a co-phasing detection method based on reinforcement learning is proposed to find a more practical co-phasing method.Secondly,the synthetic aperture imaging system model is derived based on the traditional telescope imaging principle,and the influence of co-phasing error on the imaging performance of the system is discussed from the spatial domain and frequency domain,which lays a theoretical foundation for the subsequent progress.Then,the feasibility of co-phasing sensing method using reinforcement learning is discussed.According to the idea of the optimization algorithm in the co-phase method,in order to achieve the best imaging performance of the system,it is necessary to find the mapping relationship between the influence of the co-phasing error in the process and the optimal correction action,optimize the process from the imaging performance to the control signal,and obtain the optimal policy.The residuals of the system under the optimal strategy represent the maximum correction ability for the current co-phasing error.The relevant theoretical knowledge of Markov is an effective mathematical tool for solving time series problems,and the optimization process can be modeled as a Markov decision process to solve the co-phasing problem of synthetic aperture.The Markov property of the state transition of the synthetic aperture system is demonstrated.On this basis,the compensation and correction process of the co-phasing error belongs to the Markov decision process.The optimization method of synthetic aperture system based on reinforcement learning is discussed under the framework of Markov theory.Finally,a simulation system platform of two-aperture array is built.Based on the Q-learning algorithm,we use the image definition evaluation index as the reward function,and adopt a random strategy during the strategy exploration.On this basis,the results of the co-phase detection under the point target and extended target are verified.The simulation results show that under the optimal strategy,the method can achieve fast and accurate co-phasing in different target scenarios,and at the same time has certain stability and robustness. |
