Research On High-resolution Dynamic Imaging Method Based On Fourier Ptychographic Microscopy

In the field of biological research,high resolution optical imaging with large fieldof-view(FOV)and quantitative phase imaging have always been two important research topics.The former allows us to see finer details,and extends the observed antenna into the interior of the cell,the later can quantify the phase information determined by the refractive index of the sample which makes the result more informative.Fourier ptychographic microscopy(FPM),as a new computational optical imaging method,which is different from the traditional microscope's "what you see is what you get" observation mode.This method could achieve revolutionary observation with both the above-mentioned capabilities by means of spectrum fusion.Therefore,it provides new possibilities for applications in the fields of living biological microscopic imaging and industrial inspection.However,there are still some problems in this technology,which restrict its performance.First,static disturbances such as system aberrations will cause artifacts and other phenomena in the reconstruction process.Secondly,traditional iterative algorithms have high requirements on the accuracy of the illumination wave vector,and cannot directly reconstruct the full field of view,they could only adopt the strategy of stitching several reconstructed results with small FOV,resulting in low efficiency when reconstructing a large field of view.In addition,the reconstruction process of multiple iterations of the traditional algorithm also leads to a large amount of system computation,which further reduces the reconstruction efficiency.Third,the traditional flat-panel array lighting structure will increase the spectral overlap rate of edge LEDs due to the uniform LED arrangement in the spatial domain,which will bring unnecessary information redundancy and reduce the collection efficiency of the system.Finally,the current technology cannot well cope with the ubiquitous demand for 3D observation of thick samples in the biological field.In this paper,the following research work is carried out to solve the above problems:1.To address the influence of the existence of aberrations in the FPM system on the reconstruction results,a forward imaging model embedded in the pupil recovery process is established.The reverse gradient descent algorithm combined with the second-order gradient information is used to realize the synchronous reconstruction of the pupil function and the complex amplitude of the sample,which effectively eliminates the reconstruction artifacts and blurring caused by the system aberration and improve the algorithm's robustness against system aberrations.The MSE(× 0)-3)of the reconstruction result is reduced from 23.87 to 1.05 for an aberration with a PV value of 0.32.2.Aiming at the problems that the reconstruction algorithm is sensitive to the illumination wave vector and the low speed of reconstruction,a non-iterative FPM reconstruction algorithm based on deep convolution neural network(DCNN)is proposed.By using the dataset with high “Shannon entropy” for training,the generalization ability of the network to samples could be improved.Furthermore,the network's ability in reconstructing detailed information could be improved by using composite loss function combining with spatial and frequency information.At the same time,compared with the traditional iterative algorithm,the network could significantly reduce the dependence on the accuracy of the illumination wave vector,increase the reconstructed field of view from the original 560×560 to 1800×1800,and obtain better result at the edge area in a single large FOV reconstruction,the PV value could be reduced from 0.27 to 0.08.Meanwhile,this method could reduce the reconstruction process from 12 iterations of traditional iterativebased method to 1 step,and combined with GPU acceleration technology,the reconstruction speed is shortened from 167.5s to 112.5ms,which is improved by 3 orders of magnitude and greatly improves the reconstruction efficiency of FPM technology.3.For the lighting attenuation problem caused by oblique lighting.On the basis of the hemispherical structure,a light intensity correction measure is proposed to solve the cos-square attenuation of light intensity uniformity.A method based on spectral uniform sampling is proposed,and the LEDs are designed to be arranged in concentric circles.For the extreme lighting situation with an illumination NA of 0.98,compared with the traditional array flat panel structure,the number of LEDs required is reduced from 1201 to 61(reduced 94.9%);the illumination brightness of the edge LED is increased by 637.7 times,which significantly improves the signal-to-noise ratio of dark field images and reduces the exposure time.A LED position error correction algorithm based on spectrum search is proposed,which effectively reduces the influence of the LED problem error on the reconstruction results,and the MSE(× 0)-3)of the reconstructed spectrum is reduced from 5.5 to 0.6.Based on the above achievements,a high-resolution dynamic FPM system is designed and built,which can achieve 1549)8)sub-wavelength imaging results on the basis of the overall acquisition frame rate greater than 1Hz.The imaging field of view is 550 × 5508)2,which enables high-resolution dynamic observation of tiny organelles such as vesicles and mitochondria in living cells in vitro.4.Aiming at the 3D reconstruction of thick samples by FPM system,an LED arrangement method based on Fourier diffraction theory is proposed.And combined with the hemispherical illumination structure,187 LEDs are used to achieve a maximum illumination NA of 0.98,which efficiently reduces the number of LEDs required compared to the 3D-FPM system using a flat-panel LED array.A multi-layer diffraction model combined with Rytov approximation is proposed,and the gradient descent algorithm is used to realize the three-dimensional reconstruction of the transparent sample with a thickness of 88)and the measured reconstruction resolution of 174 9)8)in the transverse direction and 524 9)8)in the longitudinal direction is realized.Combined with the quasi-Newton method,the influence of the system aberration on the 3D reconstruction is eliminated,and the MSE(× 0)-3)of the reconstruction result is reduced from 79.4 to 9.8.In order to improve the acquisition efficiency of the above system and realize the dynamic 3D reconstruction of living samples,a reconstruction algorithm for multiple LED multiplexed illumination strategy and multiple multiplexed illumination methods are designed.Simulation and experiments are also used to analyze and verify the strategy and an overall acquisition rate better than 1 Hz is achieved.This paper focuses on the reconstruction and application of FPM technology,eliminates the influence of system aberration on the reconstructed results,and greatly improves the reconstruction efficiency.Through the improving on the system,highresolution dynamic 2D and 3D reconstruction could be achieved,which further promotes the application value of FPM technology on life science and cell microobservation.