Fig. 1. Process of light field imaging

Fig. 2. Extraction of sub-aperture images from light field images

Fig. 3. Samples of light field images and particle field distribution under different particle concentrations. (a) 0.4; (b) 0.7; (c) 1.0

Fig. 4. Schematic of PRCNN structure used for reconstruction of tracer particle three-dimension distribution

Fig. 5. Schematic of the convolutional layer with residual connections

Fig. 6. Curve of cosine annealing learning rate

Fig. 7. Variation of loss function and accuracy during the training process

Fig. 8. Spatial distribution of tracer particles with a particle concentration of 0.4. (a) Actual distribution; (b) PRCNN reconstruction results; (c) SART reconstruction results

Fig. 9. Spatial distribution of tracer particles with a particle concentration of 0.7. (a) Actual distribution; (b) PRCNN reconstruction results; (c) SART reconstruction results

Fig. 10. Spatial distribution of tracer particles with a particle concentration of 1.0. (a) Actual distribution; (b) PRCNN reconstruction results; (c) SART reconstruction results

Fig. 11. Variation in reconstruction quality factor with sample particle concentration

Fig. 12. Comparison of actual particle distribution and reconstruction results on the X=50 voxel plane when the tracer particle concentration is 0.7. (a) Comparison of actual distribution and PRCNN reconstruction results; (b) comparison of actual distribution and SART reconstruction results

Fig. 13. Schematic of experimental system of cylinder wake flow field inside a horizontal square pipe

Fig. 14. Light field images of cylinder wake flow field captured in consecutive two frames in the experiment

Fig. 15. Measurement of three-dimensional velocity field based on the prediction model

Fig. 16. Measurement results of longitudinal and transverse velocities on the Z/D=0 plane. (a) Planar PIV; (b) PRCNN; (c) SART

Fig. 17. Comparison of transverse velocities on the Z/D=0 plane