Deep learning in computational imaging

Deep learning has great potential in computational imaging. We propose to use three kinds of artificial neural networks in phase imaging works. An improved U-net is used to do phase unwrapping with a new phase dataset generation method and do phase imaging in an optical microscope with Transport of Intensity Equation (TIE). And then, Y-Net and Y4-Net are used to do single-wavelength and dual-wavelength digital holographic reconstruction, respectively.