Research And Design Of Multi-scale Coding Fourier Ptychographic Microscopy

Fourier ptychographic microscopy is a newly developed computational imaging technology,which is capable of reconstructing images with a wide field of view and high resolution.However,the reconstruction based on traditional algorithms is less efficient for high calculation cost,large amount of image acquisition,and poor imaging performance.Therefore,in this work,we propose a deep learning-based multi-scale coding neural network framework for Fourier ptychographic microscopy,which improve imaging performance and reconstruction efficiency,and is robust to system errors and noise of different intensities.The research contents of this thesis are as follows:Firstly,a deep encoder-decoder network is developed to reconstruct the image.Utilizing the fitting and generating capabilities of the down-sampling and up-sampling process in the encoder-decoder network,the network can perform deeper feature recognition and better reconstruction of the image.Experiments show that the reconstructed images performance of the encoder-decoder network is better than traditional algorithms,and it is robust to system errors and noise.Secondly,we developed a residual dense connection network combining multiple technologies,including residual structure,dense connection,channel attention mechanism and sub-pixel convolution.The diamond sampling method has been used to reduce the number of acquired images.The network can expand the depth of the network,mine useful features,enhance the model ability to express and generalize,and speed up image reconstruction.Experiments show that this method reduces the number of images that need to be acquired and improves reconstruction performance and speed.Finally,in order to make the reconstruction method suitable for imaging system with multiple number of LED modes,a multi-scale multi-modal network is proposed.The network is used to reconstruct the amplitude information and phase information of the image.Experimental results show that the reconstruction method is robust to gaussian noise and it is universal to imaging systems with multiple LED quantities.