High-Precision Cross-Sensitivity Mitigation Using CNN-BiLSTM for Multiparameter Optical Sensing

Abstract

Crosstalk decoupling of multiparameters based on fiber optic sensors is crucial for high-precision detection in complex environments. The traditional sensitivity matrix method (SMM) extracts different parameters through the linear relationship between the spectral eigenvalue drift and the physical quantity to be measured. However, this scheme requires that the sensitivity responses of the different parameters be linear. To address the significant errors caused by nonlinear sensitivity in SMM, the combination of convolutional neural network and bidirectional long short-term memory (CNN-BiLSTM) model was proposed in this work. The information containing the full spectrum rather than only the peak wavelength is utilized to establish the relationship with temperature and strain. Especially when the sensitivity is nonlinear, the parameters can also be extracted accurately. Experimental results show that the deep learning-assisted approach improves the root mean square error (RMSE) of temperature and strain measurements by 9 and 44 times, respectively, compared to the SMM. This CNN-BiLSTM-based interrogation scheme may offer a novel approach to multiparameter demodulation for various sensors, significantly enhancing performance.