Research On Multi-resolution Single Pixel Image Reconstruction Algorithm

The use of structured light to activate fluorescent proteins can enable live cell imaging through saturation and depletion of non-linear SIM.Such imaging methods require multiple activations of fluorescent proteins.The strong light will damage the cells and cannot be used for live cell imaging.Therefore,single pixel imaging is used to modulate the structured light with DMD to reduce the light intensity and reduce the damage to the cells.Single pixel imaging in a weak light environment can have a better signal-to-noise ratio.Different light is modulated by DMD,and fluorescent protein is activated to collect information at the same time,and image reconstruction is performed by algorithm.Aiming at the phenomenon that single-resolution single-pixel imaging has high time complexity,the multi-resolution single-pixel imaging system is studied,and the measurement vector is constructed and applied to the multi-resolution reconstruction algorithm based on the region of interest.Perform appropriate partial sampling in the feature space to improve measurement efficiency.A single pixel multiple exposure sampling method is used to make the number of measurements much lower than the total number of image pixels.Using the optimized observation matrix can improve the reconstruction accuracy,and can obtain a high-resolution image with less sampling data.The F-spline function is introduced to construct the measurement vector and sampling matrix of single-pixel imaging.Construct F-spline measurement vector and sampling matrix based on B-spline.It is proved through experiments that the constructed sampling matrix achieves efficient image compression sampling,and the image can be reconstructed through computational imaging,and the PSNR reaches 30.5 dB.In order to reduce the damage of structural light to glial cells,a deep network was introduced to extract the region of interest,which was combined with f-spline measurement vector single-pixel imaging to achieve a multi-resolution single-pixel imaging system.Through supervised learning,deep convolutional networks learn the features of glial body images.The multiresolution region is marked on the basis of preliminary measurement.The reconstruction algorithm reconstructs and fuses the high resolution region and the low resolution region.Using glial cells to construct the test data set,the highest resolution region in the proposed single-pixel imaging algorithm PSNR reached 30.7 dB.