CNN-BiLSTM-transformer-PINN-based dual-channel loss decoupling for ultra-high sensitivity dual-core PCF-SPR sensor

Abstract

In response to the problems of channel crosstalk, insufficient sensitivity, and the loss of full spectral information and inability to separate superimposed losses in the dual-parameter photonic crystal fiber surface plasmon resonance (PCF-SPR) sensor, this paper designs a dual-core D-shaped PCF-SPR sensor and proposes an end-to-end dual-channel loss decoupling model that integrates the physical information neural network (PINN). The sensor adopts an upper and lower independently polished dual D-shaped structure. The upper surface is coated with gold film and titanium dioxide ( T i O 2 ) for refractive index detection, and the lower surface is coated with gold film and polydimethylsiloxane (PDMS) for temperature detection, achieving physical isolation of the two sensing channels. Simulation results show that the sensor achieves maximum sensitivities of 44,000 nm/RIU (within the 1.41–1.42 RIU interval) and 28.4 nm/ ∘ C (within the − 15 ∘ C to − 10 ∘ C interval), with full-range average sensitivities of 9025 nm/RIU and 5.7 nm/ ∘ C over the wide detection ranges of refractive index 1.30–1.42 RIU and temperature − 15 ∘ C to 100 ∘ C, respectively. Based on this, this paper proposes a CNN-BiLSTM-Transformer hybrid architecture to extract in parallel the local resonant features and global dispersion correlations of the spectrum, and to achieve global parameter inversion and fine decoupling of wavelength loss through the collaborative output of scalar-sequence dual heads. Test results on simulation datasets show that the model achieves R 2 values of 0.9999 for global parameter inversion of temperature and refractive index and for wavelength loss decoupling, and can accurately restore the complete spectral morphology of the two loss components. This research provides a new technical approach for high-precision multi-parameter simultaneous detection of PCF-SPR sensors, and has important application value in the fields of biomedicine and environmental monitoring.