Research on photon-level ghost imaging restoration based on deep learning

Abstract In the extreme environment, the signal of the target object is very weak, even reaching the photon level. The single-photon detection technology can detect the weak signal effectively. Ghost imaging (GI) system with single-photon detection technology has a good application in weak light imaging, but the quality of the photon-level GI is still poor. In this paper, we develop a conditional generative adversarial network (CGAN) algorithm to restore images, and the network is trained with simulation data. The results show that the CGAN algorithm has an excellent performance, and the average PSNR and SSIM of test results increase to about 20.8 dB and 0.87 respectively, proving the proposed method is feasible. It is of great significance for imaging under low photon conditions.