DCNN-Attention enhanced optical fiber glucose sensor with ultra-high sensitivity based on the optical Vernier effect

Description

DCNN-Attention core code: The proposed DCNN-Attention network is an end-to-end demodulation model designed for the high-precision regression of high-dimensional optical fiber sensing spectra. It seamlessly integrates sequential deep convolutional blocks with Convolutional Block Attention Modules (CBAM) and a global context-weighting mechanism to effectively capture and enhance critical feature representations, thereby achieving precise target value regression based on 4001-dimensional optical fiber spectral data. The supplementary materials referred to in the manuscript consist of four subsections: Section S1 presents the detailed configuration parameters and data dimensions for the CNN, DCNN, and DCNN-Attention models. Sections S2–S4 provide the complete Vernier spectra across all experimental concentrations, the error histograms for the CNN and DCNN models, and the spectral comparisons that validate the robustness and generalization capability of the DCNN-Attention model discussed in Section 3.5. Video Intelligent Demodulation Process: This video demonstrates the complete experimental workflow of the proposed DCNN-Attention model, capturing the training phase on a remote workstation, the deployment of the optimal model on a local PC for real-time spectral demodulation, and an illustrative example of transfer learning. It offers a practical perspective on how deep learning architectures are implemented in real-world engineering scenarios for advanced optical fiber sensing demodulation.

Institutions

• Tianjin University of Technology

  Tianjin, Tianjin

Categories

Funders

• Tianjin Science and Technology Plan Project

  Grant ID: 24JCZXJC00510

Licence