Fig. 1. Physics-informed network architecture for 3D particle field reconstruction in light field microscopy. (a) Model pre-training process; (b) model fine-tuning process

Fig. 2. Light field microscopic imaging process

Fig. 3. Schematic diagram of reconstruction network

Fig. 4. Schematic diagram of convolution calculation process

Fig. 5. Variation of loss function during the correction process

Fig. 6. Particle light intensity distribution reconstructed using different models. (a) Ground truth of particle light intensity distribution; (b) light intensity distribution reconstructed using U-Net model; (c) light intensity distribution reconstructed using PIDLR model

Fig. 7. Reconstructed light intensity distribution of a single particle (region ③ in Fig. 6). (a) x-direction light intensity distribution reconstructed using U-Net model and PIDLR model; (b) z-direction light intensity distribution reconstructed using U-Net model and PIDLR model

Fig. 8. Reconstruction quality at different particle concentrations

Fig. 9. Reconstruction results of U-Net model and PIDLR model under different SNRs. (a) Light intensity distribution; (b) reconstruction quality

Fig. 10. Reconstructed examples of particle fields with particle concentrations of 0.3 ppm and 1.3 ppm using U-Net model and PIDLR model

Fig. 11. Experimental setup for microscale flow measurement. (a) Schematic diagram of experimental setup; (b) Y-typed microfluidic chip

Fig. 12. Three-dimensional velocity field of particle field reconstruction using PIDLR model

Fig. 13. Comparisons of measured and theoretical velocities at different x positions over xoy plane of z=50 μm. (a) x=140 μm; (b) x=170 μm; (c) x=200 μm