Full-Range Ultrasensitive Fiber-Optic Hydrogen Detection via Twin Vernier Amplification and Lightweight CNN.

The sensor incorporates a polydimethylsiloxane (polydimethylsiloxane)-filled temperature reference cavity and a polydimethylsiloxane/Pd-WO3 hydrogen-sensitive cavity on a single fiber, enabling intrinsic temperature compensation and humidity-insensitive detection without external reference devices.

To resolve the wavelength ambiguity imposed by the free spectral range (free spectral range), we combine physics-guided spectral data augmentation with a lightweight dual-branch fusion convolutional neural network regression model to achieve end-to-end 0-100% volume fraction hydrogen concentration inversion, with an R2 of 0.999443 and a root mean square error (RMSE) of 0.682%.