In this paper, to the best of our knowledge, we first realize imaging of an object passing through multimode fiber (MMF) by using upconversion imaging technique and deep learning. It is shown that as infrared object passes through MMF and/or scattering medium, the infrared speckle is obtained, which cannot be detected by CCD working in visible range. Using upconversion imaging technique based on second harmonic generation, we can shift the infrared speckle pattern to the visible one, which can be recorded by CCD working in visible range. We employ a network model, named pix2pix, which is built based on the generative adversarial network (GAN), to realize image reconstruction. It is found that the trained network can reconstruct image from the visible speckle with high quality, in which the visible speckle is obtained by the upconversed imaging technique. The combination of the upconversion imaging and deep learning is found to be feasible for extending the imaging wavelength and recovering images from speckle with high fidelity.
Read full abstract