Research On Single Fiber Local Speckle Imaging Technology Based On Deep Learning

Multimode fibers(MMF)are widely used for image transmission and communication due to their high capacity and high secrecy.They can be used for imaging in confined environments in industrial and medical applications because of their slim size and high resolution.In MMF,the transmission state is highly variable due to hybrid effects such as coupling,interference,and dispersion between modes,leading to a highly random granular speckle distribution of the MMF output light field that cannot be directly imaged and requires the speckled light field to capture the original image.As a result,deterministically resolving the MMF end-to-end optical field mapping relationship is crucial.Because of the MMF transmission process’ s mode mixing effect and noise interference,MMF speckles frequently contain a lot of redundant information,resulting in a greatly reduced imaging efficiency and increased algorithm complexity.In order to achieve high-fidelity reconstruction using a deep learning neural network architecture and improve the quality and speed of speckled reconstructed images.These are the two main areas of research:1.The point spread function transmission theory of MMF transmission was analyzed,an optical path of MMF image transmission was built,the image was loaded into the optical path by spatial light modulator(SLM),and CCD camera images of the front and back ends of MMF were taken and acquired.To investigate the redundancy characteristics of the speckled light field,the acquired global speckled light field was split up local position and local area,and a dataset of 39 different local speckles was created.Further,two random pixel reorganization speckled datasets were created to investigate the randomness of the speckled particle distribution.2.The Swin-Unet was used to recover speckled images,and its properties of global association and local attention to speckled features were investigated.Image reconstruction was trained and tested for 41 types of local speckles,and comparing the recovery effects with those of global speckles.Experimental results demonstrate the local speckled light field has the ability to recover the image,where the larger the speckled area and closer to the center,there is more useful feature information.Also,the effective feature information distribution of the speckled light field is random,and randomly speckled can also recover the image well.Therefore,image reconstruction using local speckle can simplify the network and reduce computing resources.