A Deep Learning-Based Fusion Framework for Temperature-Aware Relative Humidity Sensing Using Dual Fiber-Optic Sensors

This work presents a fiber-optic sensor fusion framework for temperature-aware relative humidity sensing. The proposed system integrates a micro-loop resonator (MLR) coated with hydroxyethyl cellulose/poly(vinylidene fluoride) (HEC/PVDF) and a D-shaped fiber coated with Polydimethylsiloxane (PDMS). The two sensors were experimentally characterized, achieving a 0.076 dBm/%RH sensitivity to relative humidity for the MLR and a 0.045 dB/°C response to temperature for the D-shaped fiber. A multimodal dataset was produced from the MLR's transmission spectrum and the D-shaped fiber's power output to train a deep learning fusion model combining convolutional neural networks (CNN) and a multilayer perceptron (MLP). This approach was selected to exploit the full spectral characteristics of the MLR rather than single-peak features, enabling robust compensation of temperature-induced cross-sensitivity. The fusion model was trained using a leave-one-temperature-out approach and achieved a mean absolute error of 0.499 ± 0.173 %RH. The results demonstrate that deep-learning-based sensor fusion effectively mitigates temperature cross-sensitivity and improves the robustness of relative humidity sensing.