Fig. 1. Principle diagram of FET reconstruction

Fig. 2. Structure of the 3D-2D hybrid network

Fig. 3. 3D-2D tandem structure

Fig. 4. Three-dimensional structure of the fire simulation field

Fig. 5. Diagram of the experimental setup

Fig. 6. Projection images acquired by 12 camera simulations

Fig. 7. Network loss curves for different structural parameters. (a) Different sampling layers; (b) different numbers of channels; (c) different convolution kernels

Fig. 8. 3D reconstruction results of ART and HNN

Fig. 9. Comparison of ART and HNN reconstruction results. (a) SSIM; (b) RMSE; (c) CORR; (d) PSNR

Fig. 10. Two-dimensional qualitative comparison of ART and HNN reconstruction results.

Fig. 11. Comparison of training results between VT-Net2 and HNN on the validation set

Fig. 12. Projection images of flames in 12 directions from experimental measurements. (a) Two flames; (b) three flames

Fig. 13. Projection images after downsampling. (a) Two flames; (b) three flames

Fig. 14. Experimental flame reconstruction slices. (a) Two flames, z=12; (b) three flames, z=14

Fig. 15. Comparison of the 3D visualization of the reconstruction experiment results between ART and HNN. (a) 3D reconstruction of the 2nd moment of two flames; (b) 3D reconstruction of the 186th moment of two flames; (c) 3D reconstruction of the 50th moment of three flames; (d) 3D reconstruction of the 100th moment of three flames

Fig. 16. Comparison of reconstruction experiment results of two flames between ART and HNN. (a) SSIM; (b) RMSE; (c) CORR; (d) PSNR

Fig. 17. Comparison of reconstruction experiment results of three flames between ART and HNN. (a) SSIM; (b) RMSE; (c) CORR; (d) PSNR

Fig. 18. Visual comparison of two-flame reconstruction experiment results between ART and HNN. (a) z=40; (b) z=70

Fig. 19. Visual comparison of three-flame reconstruction experiment results between ART and HNN. (a) z=40; (b) z=70