Deep learning for acousto-optic sensing

Acousto-optic sensing offers a unique capability for sound measurement. Thanks to its non-contact nature and the use of advanced optical instrumentation, acousto-optic sensing enables instantaneous visualization of two-dimensional sound fields with resolutions orders of magnitude higher than those of microphone arrays. One major challenge in acousto-optic sensing, however, is its intrinsically low signal-to-noise ratio. Because the physical interaction between light and sound is inherently weak, the measurement data are often affected by noise. To address this issue, deep learning-based methods have recently been applied to acousto-optic sensing data, demonstrating significant improvements in tasks such as denoising, segmentation, extrapolation, and up-sampling. This talk will cover these recent developments and discuss the remaining challenges.