Research On Optical System Based On Aberration Compensation Network

Multimode fiber has a small core diameter and can transmit multiple spatial light modes at the same time.Its application in medical endoscope imaging is expected to solve the problems of large probe size and low resolution of traditional optical fiber endoscope,so it has the potential to broaden the microscopic channels of medical endoscopes.As one of the multimode fiber imaging techniques,transmission matrix method has been widely concerned.Conventional transmission matrix acquisition methods require precise calibration,repeated measurements,high data computation,and any bending or perturbation of the fiber can lead to changes in the transmission matrix,resulting in serious impairment of imaging quality,which greatly limits the application of imaging using multimode fibers.Therefore,fast and accurate estimation of the transmission matrix has been a thorny problem with great significance.The aberration compensation network is a lightweight neural network with fast convergence,making it possible to estimate the transmission matrix quickly and accurately.The main work of this paper tries to overcome the defects of the transmission matrix acquisition method,hoping to solve the transmission matrix more quickly and accurately based on the aberration compensation network,to compensate the perturbation of the optical fiber in the optical system more accurately,and to provide a new imaging pathway for its later application in multimode fiber imaging.The innovation points and main work of this paper are as follows:1.An aberration compensation network improvement scheme with the pursuit of diagonalization of the modal transmission matrix as the optimization objective is proposed.Aiming at the defect that the compensation network does not add a penalty term to the non-diagonal elements of the transmission matrix,this paper proposes an improved scheme of the compensation network to pursue the optimization objective of the modal transmission matrix diagonalization.The simulation results show that the diagonal energy ratio of the compensation network to solve the mode domain transmission matrix increases from 92%to 96%after the improvement.2.Verified that the improved aberration compensation network can more accurately compensate for fiber perturbations,specifically in terms of transmission matrix diagonalization and pattern image recovery.A modeling scheme to realize fiber perturbation by acting optical aberrations in fiber modes is proposed.By comparing the output results of the improved aberration compensation network before and after,it is verified that the improved aberration compensation network has a great improvement in mode image recovery and transmission matrix diagonalization.The average Pearson correlation coefficient between the recovered and simulated modes of the improved aberration compensation network is increased from 0.73 to 0.95.The experimental results well demonstrate that the improved aberration compensation network can more accurately.The experimental results demonstrate that the improved aberration compensation network can more accurately compensate for the perturbation of the fiber.3.Verified that the mode of aberration compensation network recovery is not invariant in fiber transmission.The modal transmission matrix,theoretically represented by propagationinvariant modes base,is a diagonal matrix,and applying this transmission matrix to a multimode fiber optic imaging system can enhance the imaging quality.Therefore,it is possible to test whether the mode recovered by the aberration compensation network remains transport invariant in the fiber by matrix multiplication of the fiber mode with its corresponding pixel-domain transport matrix.The experimental results show that the recovered mode of the network has a low Pearson correlation coefficient with its output mode through the corresponding pixel-domain transmission matrix,indicating that the recovered mode of the network is not invariant in multimode fiber,because two identical models of the aberration compensation network are trained on two identical straight fiber modes at the same time,resulting in the inconsistency of the recovered mode matrix of the two models after the network optimization is completed.This experimental result provides a reference for further improvement scheme of the network.