Influence of the spatial coherence of the illuminating laser on the imaging through scattering media via deep learning

Deep learning -based imaging through scattering medium has been increasing interests in recent years. In general, the object is illuminated by a coherent laser, and the speckle is formed as the light passing through the scattering medium. The neural network is training by using a lot of object -speckle pairs, and the trained neural network can reconstruct imaging from the speckle. In this paper, we study the more general case that the object is illuminated by partially coherent laser. The influence of the spatial coherence of the illuminating laser on the imaging performance is studied. We employ generative adversarial network (GAN) for reconstructing the imaging in the case that the object is illuminated by laser with controllable spatial coherence. It is found that even the spatial coherence is quite low, the imaging performance is good. Moreover, we will study the situation at which only very small part of the speckle is collected. It is demonstrated that based on this very small part of the speckle, the performance of reconstructing image remains quite good.