Fig. 1. The recording schematic of THz in-line digital hologram

Fig. 2. Neural network algorithm based on PhysenNet. (a) The flowchart; (b) the schematic diagram of the U-Net

Fig. 3. The simulations of in-line digital holography based on the PhysenNet method. (a) (b) The amplitude and phase distributions of the simulated samples; (c) the simulated hologram; (d) the results reconstructed by PhysenNet with different iterations; (e) the loss function curve

Fig. 4. Comparison of numerical simulation results of different phase retrieval methods. (a)-(e) and (f)-(j) The reconstructed amplitude and phase distribution by the ASP, ER, IDPR-RI, CCTV, and PhysenNet method, respectively

Fig. 5. Schematic of continuous waves THz in-line digital holography

Fig. 6. The amplitude distributions of siemens star obtained by different phase retrieval algorithms. (a)-(c) Physical images, holograms, and normalized holograms of samples; (d)-(h) the amplitude distributions by the ASP, ER, IDPR-RI, CCTV, and PhysenNet method, respectively

Fig. 7. The reconstructed results of the cicada wing by different phase retrieval algorithms. (a) the optical photo of the sample; (f) the normalized hologram; (b)-(e) and (g)-(j) the amplitude and phase distributions by the ER, IDPR-RI, CCTV, and PhysenNet method, respectively

Fig. 8. Comparison of the reconstructed results of a PS foam sphere. (a) The optical photo of the sample; (b) the normalized hologram; (c1)-‍(g1) and (c2)-(g2) the amplitude and phase distributions by the ASP, the ER, the IDPR-RI, the CCTV, and the PhysenNet method, respectively