Fig. 1. Principle of the MTP-MZI sensor

Fig. 2. Light field simulation of the MTP-MZI sensor

Fig. 3. Schematic diagram of MTP-MZI sensor fabrication device

Fig. 4. Schematic diagrams of preparation process for MTP-MZI sensor. (a) Schematic diagram of two optical fibers in the fusion splicer; (b) schematic diagram of the welding region after arc discharge; (c) microscopic image of the MTP-MZI sensor

Fig. 5. Simulation and actual spectra of the sensor. (a) Simulation and actual transmission spectra of the sensor; (b) actual spatial frequency spectrum

Fig. 6. Test device for polarization sensitivity

Fig. 7. GAF image encoding of one-dimensional spectral sampling sequence. (a) Transmission spectrum; (b) downsampling result; (c) polar coordinate mapping; (d) GAF image

Fig. 8. Structure of AlexNet

Fig. 9. Curvature detection sensor experiment setup

Fig. 10. Performances of MTP-MZI sensor under different curvatures. (a) Transmission spectra of sensor under different curvatures; (b) linear fitting between peak intensity and curvature

Fig. 11. Transmission spectra of MTP-MZI sensor under larger curvature variation

Fig. 12. Variation in loss of training set and coefficient of determination of test set with epoch for MTP-MZI sensor

Fig. 13. Comparison of predicted and applied curvatures on the validation set for MTP-MZI sensor

Fig. 14. Schematic diagram of the in-line MZI sensor structure

Fig. 15. Transmission spectra of in-line MZI sensors with different CLF lengths

Fig. 16. Performances of in-line MZI sensor under different curvatures. (a) Transmission spectra of sensor under different curvatures; (b) linear fitting between peak intensity and curvature

Fig. 17. Transmission spectra of in-line MZI sensor under larger curvature range

Fig. 18. Variation in loss of training set and coefficient of determination of test set with epoch for in-line MZI sensor

Fig. 19. Comparison of predicted and applied curvatures on the validation set for in-line MZI sensor

Fig. 20. Prediction performances of different CNN networks on sensor curvature

Fig. 21. Average training time per epoch of different CNN networks