Fig. 1. Bayesian deep neural network architecture for phase gradient calculation

Fig. 2. Phase maps generation process. (a) Pseudo-random numbers; (b) phase maps using Lagrange extrapolation; (c) phase gradient label; (d) wrapped phase map with speckle noise

Fig. 3. The prediction process of phase gradient using Bayesian neural network

Fig. 4. Phase gradient estimated using different methods. (a) Phase maps with different noise levels; phase gradient results obtained by (b) theorectical calculation; (c) vector method and (d) Bayesian neural network; (e) BNN model uncertainty; (f) error distributions of phase gradient by using BNN model

Fig. 5. (a) Schematic diagram and (b) photograph of line-field spectral-domain OCT system

Fig. 6. Experimental results of silicone film deformation. (a) Wrapped phase-difference map; (b) phase gradient estimated using vector method; (c) phase gradient estimated using BNN; (d) BNN model uncertainty

Fig. 7. Experimental results of phase decorrelation. (a)-(b) Wrapped phase-difference maps corresponding to the loading 12 μm and 14 μm, respectively; (c)-(d) results of phase gradient estimated using BNN; (e)-(f) BNN model uncertainty