Accuracy enhancement of multi-core fiber shape sensing by a deep learning-based model with pretraining and transfer learning strategies

We propose a shape-space coordinate prediction model for multi-core fiber Bragg grating (MCFBG) sensors, which integrates pretraining and transfer learning strategies with deep learning architectures. The model establishes an end-to-end mapping relationship from the center wavelength data of MCFBGs to their corresponding shape-space coordinates, which improves the accuracy of MCFBG-based shape sensing while reducing the amount of training data required in experiments. Results show that the best-performing model achieves a median terminal point error with a relative error as low as 0.76%. The proposed method holds strong potential for high-precision shape sensing applications.