Research On Fourier Ptychography Imaging Algorithm Based On Convolutional Neural Network

In optical imaging systems,imaging resolution and field of view are a pair of contradictions that are difficult to reconcile,and Fourier ptychography imaging can overcome this contradiction and is a computational imaging technology that simultaneously achieves high imaging resolution and large field of view.The traditional Fourier ptychography imaging algorithm to reconstruct high-quality images needs to acquire a large number of low-resolution images,resulting in low sampling efficiency of the imaging system.Moreover,the convergence of the algorithm needs to go through many iterations,the reconstruction speed of the algorithm is slow,and the time cost is high.In this paper,the convolutional neural network in deep learning is combined with Fourier ptychography imaging,and a reconstruction algorithm based on convolutional neural network is proposed.By reducing the number of low-resolution images required for reconstruction,the sampling of Fourier ptychography imaging is improved.and the proposed algorithm has a fast reconstruction speed,which can improve the reconstruction efficiency.The main research contents of this paper are as follows:A Fourier ptychography imaging based on multi-scale feature fusion network is proposed,which combines the feature pyramid network in convolutional neural network with Fourier ptychography imaging.The feature pyramid network is composed of an encoder and a decoder.The encoder part can extract the feature information corresponding to different frequencies of the imaged objects from the low-resolution image set,and introduces the dense connection and channel attention mechanism to improve the network’s ability to feature information.screening ability.The decoder part can restore the feature map size and fuse feature information through bicubic interpolation upsampling and convolution layers.Experiments show that,compared with the traditional reconstruction algorithm,the Fourier ptychography reconstruction algorithm based on the multi-scale feature fusion network not only has a faster reconstruction speed,but also reduces the number of images required for reconstruction by reducing the aperture overlap ratio and improves the sampling efficiency.And it has strong robustness to Gaussian noise.The Fourier ptychography reconstruction algorithm based on the multi-scale feature fusion network can greatly reduce the aperture overlap rate and the number of samples,and improve the sampling efficiency.However,in the macro Fourier ptychography imaging system,the sampling is limited by the aperture overlap ratio,and scanning sampling is still required.Therefore,the proposed convolutional neural network is further improved.The original convolutional neural network is used as the generator,and two discriminators are added.The generator and the discriminator together form a generative adversarial network.The generative adversarial network can still reconstruct the Fourier ptychography imaging when the sub-aperture on spectrum has no overlap ratio,and the use of dual discriminators effectively improves the quality of the reconstruction.Experiments with simulated data and actual data show that it is feasible to use convolutional neural network to realize fast Fourier ptychography imaging with no overlap of apertures.