光纤
深度学习
人工智能
计算机科学
纤维
人工神经网络
光学
光纤传感器
信号处理
材料科学
电子工程
光学滤波器
集成光学
模式识别(心理学)
干扰(通信)
深层神经网络
工程类
作者
Lixia Li,Bin Hang,Minglong He,X Y Zhao,Shuoyang Yu,Yufang Liu
标识
DOI:10.1109/jlt.2026.3736469
摘要
Precise and real-time measurement of solution refractive index (RI) is essential for environmental monitoring, food safety inspection, and clinical biosensing. In this work, a silver-based fiber optic surface plasmon resonance (SPR) sensor is investigated for RI detection. Conventional SPR demodulation methods generally rely on limited spectral descriptors and therefore cannot fully exploit the discriminative information distributed throughout the broadband resonance spectrum. Moreover, existing CNN- or recurrent-based approaches tend to emphasize either local spectral morphology or sequential wavelength dependencies, leaving their complementary information insufficiently integrated. To address this gap, a convolutional neural network-bidirectional gated recurrent unit-attention (CNN-BiGRU-Attention) framework is proposed for full-spectrum analysis. The CNN extracts multi-scale local spectral features, the BiGRU captures bidirectional long-range wavelength dependencies, and the attention mechanism adaptively emphasizes RI-sensitive spectral regions while suppressing less informative responses. By jointly modeling local spectral morphology, global wavelength correlations, and region-specific spectral saliency, the proposed framework provides a unified representation of the complementary RI information encoded throughout the complete SPR spectrum. Experiments conducted on five RI standards spanning 1.333–1.373 RIU yielded a coefficient of determination (R2) of 0.9950, a root mean square error (RMSE) of 0.0017 RIU, and a mean absolute error (MAE) of 0.0013 RIU. These results demonstrate the strong potential of the proposed SPR sensing framework for real-time environmental, biomedical, and industrial sensing applications.
科研通智能强力驱动
Strongly Powered by AbleSci AI