Neural Surrogate Modeling of Ultrasonic Wave Propagation in Heterogeneous Media
The LU-NO method integrates Fourier Neural Operators and local-integral resampling into a U-Net architecture to model ultrasonic wave propagation through complex, heterogeneous media. This approach successfully captures essential signal features like reflections while maintaining robustness to input noise and performing accurately across different spatial scales, offering a computationally efficient alternative to traditional solvers.