The Research On The Characteristics Of Image Transmission Of Multimode Fiber

Optical fiber imaging has been a research hotspot in optical field.With the development of new optical fiber technology,optical fiber gradually plays a vital role in medical endoscope,industrial endoscope and other equipment as an image transmission tool.Especially after the appearance of optical fiber bundles,the application of optical fiber imaging increases rapidly,but it is difficult to meet the requirements of both the radius and the resolution of optical fiber bundles.Multimode fibers can transmit hundreds or even thousands of modes in a single fiber core,and the pixel density is 1-2 orders of magnitude higher than that of fiber bundles,which meets the requirements of high-precision,small-diameter and miniaturization of endoscopes and other devices.However,because multimode fiber is a highly sensitive and strongly scattering medium,not only random speckle will be formed at the output end,but also it is vulnerable to disturbance.Even a slight vibration or temperature change will cause the transmission matrix to change greatly,and it is difficult to restore the real image at the far end,which hinders the application and development of multimode fiber as an imaging tool.As an imaging tool such as an endoscope,the image transmission quality of multimode fiber will directly affect the use of the equipment,so the understanding of its image transmission characteristics will help us to implement it into practical applications better and faster.Wavefront shaping and transmission matrix are traditional imaging methods of multimode fiber based on experimental measurement,the equipment is complex and the stability is poor.In recent years,deep learning has been developed rapidly with the progress of software and hardware,and it has been successfully used in auxiliary imaging of multimode fiber,showing great potential.In this paper,the image transmission characteristics of multimode fiber are studied by using neural network method.The innovation and main work of the paper are as follows:1.Neural network is used to simulate the image transmission behavior of multimode fiber,and a method that can characterize the nonlinearity of multimode fiber is explored.Assuming that the neural network can completely fit the image transmission behavior of multimode fiber,by using the input-output data of MMF to train the neural network,using the one-side inhibition characteristic of ReLU nonlinear activation function,namely "necrosis phenomenon",to control its threshold Δ to affect the nonlinearity of the whole neural network,which is represented as the changing trend of the number of necrosis neurons,the nonlinearity of MMF is characterized by this method.It is found that there is a part of MMF that can be linearly characterized,and there is indeed a part that cannot be linearly characterized.2.By simulating multimode fibers with different number of modes,it is verified that there is indeed a part of multimode fibers that can be linearly approximated.When the nonlinearity capability increases,multimode fibers can no longer be linearly approximated.The number of modes is used to characterize the nonlinearity capability of MMF.By simulating MMF with different number of modes to obtain the corresponding input-output pairs and make the data set to train the neural network corresponding to the parameter MMF,the approximate linear model of the pre-trained model at each mode number can no longer be linearly approximated.3.Exploring the relationship between the resolvable image complexity and the number of fiber modes.The method of simulating real optical fiber brings great convenience to the experiment,in the above experiment,natural images are also input to simulated optical fibers with different number of modes,and at the same time,it is also an inquiry into the relationship between the number of optical fiber modes and the complexity of the resolvable image,a series of experiments show that the effect of image reconstruction using the data of optical fibers with large number of modes is better,which shows that optical fibers with large number of modes indeed have better resolution ability to the image.4.Exploring a kind of anti-bending compressive sensing fiber optic imaging bundle for ghost imaging.The compressive sensing algorithm is introduced into the fiber optic imaging bundle to reduce the number of data sampling in the imaging process,improve the robustness of the imaging system against external disturbances,improve the problem of pixel dispersion existing in the traditional fiber optic imaging bundle,and make a forward-looking exploration for realizing the imaging scheme with high resolution and low complexity.