An Iterative Neural Network for Imaging Multi-layer Targets in Single-photon Applications

Single-photon imaging systems have been widely used to obtain three-dimensional information. In real applications, the same pixel will receive return signals from targets at different depths, which puts higher requirements on 3D reconstruction algorithms. However, existing approaches have limited restoring capabilities in conditions with low signal-to-noise ratio (SBR) and a few signal photons. Hence, we propose an end-to-end neural network model, denoted as MTSR-Net. By adopting an iterative architecture and introducing a Gaussian window module, we capture physical priors in single-photon data into the algorithm. Experimental results prove that our method has better reconstruction quality than existing methods under different SBRs.