High-resolution Ptychographical Iterative Engine For Solving The Limitation Of Numerical Aperture And Pixel Size

The ptychography is a newly developed lensless imaging method,which does not require complex optical systems,and the imaging resolution is not limited by optical lenses.Ptychography has the advantages of large field of view,no aberration,extended imaging range and long working distance,and has been widely used in the research of visible light imaging,X-ray imaging and electron beam imaging.However,the imaging resolution of ptychography is limited by the numerical aperture of the Charge Coupled Device（CCD）and the pixel size.When the numerical aperture is limited,part of the high-frequency information is easily lost when the acquired spot is close to the edge of the CCD target surface.In addition,when the pixel size is large,some detailed high-frequency information will be lost,and the loss of high-frequency information in the above two states will reduce the imaging resolution.The core of this research is to solve the limitation of ptychography resolution by numerical aperture and pixel size.The specific research contents are as follows:1.High-resolution ptychography method based on extrapolation and interpolation:This method can simultaneously deal with the problem that the resolution of ptychography is limited by the numerical aperture and pixel size of CCD.Firstly,using the redundancy and connectedness between the acquired diffraction images,the higher-order diffraction information can be obtained outside the CCD aperture by iterative extrapolation to the higher-order diffraction outside the CCD boundary.The numerical aperture limitation problem is solved.After that,the reconstruction of the extrapolation method is combined with the interpolation method.The pixel number of the reconstruction of the extrapolation method is extended,and the sampling interval of iterative calculation is reduced by pixel interpolation,which can solve the problem of limited pixel size.The image resolution of the reconstructed USAF 1951 can reach 143.7lp/mm.2.High-resolution ptychography method based on generative adversarial networks:to further improve the ptychography resolution and reduce the time required for iterative operations,a multi-weight loss function generative adversarial network is introduced into ptychography.First,the simulated e PIE algorithm is used to establish the data pairs required for the multi-weight loss function GAN training,and the multi-weight loss function is used to update the generator and discriminator during the network training.The multi-weight loss function is the weighted sum of mean square error（MSE）loss,feature map loss and adversarial loss,and the balanced processing of reconstructed image pixels and visual level is achieved by setting reasonable weights.After using the extrapolation method to supplement the higher-order diffraction information lost due to the limited CCD target surface,the resulting reconstructed image is fed into the trained generative network to expand the number of reconstructed image pixels,reduce the pixel size,quickly solve the problem of ptychography resolution limited by the CCD pixel size,and improve the imaging resolution and reconstruction efficiency.The image resolution of the USAF 1951 reconstructed by our proposed method can reach 161.3lp/mm,and the reconstruction efficiency can be improved by 1.1 times compared with extrapolation first and then interpolation,and the reconstruction efficiency can be improved by 3.75 times compared with super-resolution ptychographical iterative engine.3.High-resolution ptychography method based on sample-free training:convolutional neural networks with sample-free training mode is introduced into ptychography,the image reconstructed by extrapolation method is used as the high-resolution image I^HRfor network training,the image reconstructed by extrapolation method combined with the pixel binning method is used as the low-resolution image I^LR,the pairwise input convolutional neural network is randomly initialized,the network learns the mapping relationship between two images through iteration,and the super-resolution model of the image is obtained after iteration.The network learns the mapping relationship between the two images through iteration,and the super-resolution model of the image is obtained after the completion of the iteration,and the I^HR is input to the super-resolution model for reconstruction,which can solve the problem that the resolution of ptychography is limited by the CCD pixel size and improve the resolution of ptychography.