Reproduce “Continuous subsurface property retrieval from sparse radar observations using physics informed neural networks” on OpenResearch

About this paper

Ishfaq Aziz, Mohamad Alipour

The research from the University of Illinois Urbana-Champaign developed a physics-informed neural network (PINN) framework that reconstructs continuous subsurface dielectric permittivity profiles from sparse radar measurements. This approach accurately mapped complex material distributions in both synthetic and real-world experiments, outperforming traditional discrete-layer inversion methods by eliminating prior assumptions and achieving high data efficiency.

Key insights

• PINNs can effectively perform continuous subsurface property inversion, moving beyond discrete layer assumptions, even with sparse radar observations, by leveraging physical laws as a strong regularizer.
• Strategic placement of a minimal number of radar receivers (e.g., 3-6) is crucial for accurate permittivity profile reconstruction, especially when covering material boundaries.
• The framework can implicitly learn transmitted radar pulse characteristics, eliminating the need for explicit prior model calibration required by traditional wave inversion methods.

Summary by alphaXiv.