Light Control And Computational Imaging Through Scattering Media By Deep Learning

When the laser passes through the scattering media,the wavefront undergoes severe distortion and produces laser speckles around the scattering media,affecting the imaging resolution,which seriously restricts the rapid development of biomedical imaging,atmospheric remote sensing,laser therapy,optical fiber communication and holographic imaging.Therefore,it is of great significance in the field of optics to study the recovery of speckle and how to regulate the light field in the scattering media.Based on the computational holographic imaging system through scattering media,a neural network framework is proposed by the thesis,and on this basis,the light regulation and imaging in the scattering media are studied in depth.The main work of this thesis are as follows:1.The transmission process of light in the scattering media and the formation principle of speckle are analyzed,a computational holographic imaging system through scattering media is built,and the automatic batch loading and acquiring pictures is realized by the spatial light modulator and the API interface related to the CMOS camera for MATLAB programming,which reduces manual operation and provides a large amount of data for the dataset of the neural network.2.A new neural network structure is proposed.In the thesis,combined with the Unet network structure and residual block,the Res Unet network is proposed,which is more conducive to the extraction of image features.The BN layer is added before the activation function of the residual block,which greatly improves the convergence speed of the loss function and the stability of the network model.3.Based on the computational holographic imaging system through scattering media and the Res Unet network structure,the study of speckle recovery experiments is carried out.The experiment is divided into two sub-experiments.The first experiment is to explore the ability of speckle to restore the original picture.In order to increase the generalization ability of the network,the experiment uses the block binary random picture-speckle pairs as a data set,and for the test set,uses homemade pixel picturespeckle pairs.The results show that the neural network can recover the speckle of the pixel picture,and the finer the random picture blocks,the more complex image can be recovered by speckle.The second experiment is to explore the ability of speckle to restore hologram.This dataset uses holograms and speckles corresponding to MNIST handwritten numbers.The results show that the network can recover simple holograms.4.Based on the computational holographic imaging system through scattering media and Res Unet network structure,the regulation of light transmission in the scattering media is studied.The effects of single-point focusing and multi-point focusing are explored under the conditions of different blocks of random pictures and different data sets.The results show that the regulation of light transmission in the scattering media can be realized based on deep learning.A single focus and multiple focus at the target location can be generated after the scattering media,but in the case of thinner blocks of random pictures,the focusing effect is weaker due to the poor training effect of the neural network.