Whispering-gallery-mode sensing via physics-informed machine learning

Optical whispering-gallery-mode(WGM) resonators enable ultra-sensitive sensing by tracking resonance shifts in resonance spectra. However, reliably extracting changes in resonances from complex, evolving spectra remains a challenge that limits scalability and multi-parameter sensing. Here, we present a machine learning framework that estimates resonance shifts using spectral representations trained with physics-guided augmentations. The resulting spectral representation yields robust physical signatures for downstream sensing tasks. We evaluate our machine learning approach on temperature sensing using WGM resonators, where it outperforms traditional signal processing and existing machine learning approaches in sensing performance for both interpolation and extrapolation of temperature estimation. These results suggest physics-informed machine learning can make WGM spectral sensing more reliable across heterogeneous conditions and devices.