GNGTS 2021 - Atti del 39° Convegno Nazionale

399 GNGTS 2021 S essione 3.1 Employing 3D convolution kernels The introduction of 3D convolutions in the multi-res U-Net allows to exploit the self-similarity and the correlations between different time slices making the network more capable of capturing useful features in the data. Thismoremeaningful prior results in significantly higher performances, leading to a SNR of 18 . 5dB. The reconstruction results obtained through this architecture are shown in Figure 2. Imposing time slice symmetry Finally, we introduce a physics-related constraint to impose reciprocity between source- receiver pairs. The output of the network is averaged with its own transpose, forcing the time slice to be symmetric. Results are shown in Figure 3; thanks to this further constraint, we achieve 21 . 8dB of SNR.