Tomographic Reconstruction of Sound Field From Optical Projections Using Physics-Informed Neural Networks

Physics-informed neural networks (PINNs) are widely used as a framework for approximating the solutions of PDEs. Previous studies have shown that PINNs efficiently approx-imate arbitrary continuous functions from high-dimensional data and are suitable for solving multi-physics problems. Op-timization of continuous functions with neural networks has proven to be effective in representing fields and c an bean innovative approach to the challenges faced by optical sound measurements. Optical sound measurements enable contact-less observation of the spatial distribution of sound. Because the observed quantity is line integral of the sound pressure on the optical path, reconstruction is necessary to determine the sound pressure at each point in the three-dimensional field. In this paper, we introduce line integral loss between the opti-cally measured sound-field projection and the line integral of the neural field and propose a learning-based reconstruction method using the model constrained by Helmholz equation. The experimental results show that our method outperforms a filtered back projection, a classic tomography method.