Abstract
In this paper, a novel deep learning characterization approach is proposed and demonstrated experimentally using a pressure sensor based on a Sagnac interferometer realized by a side-hole fiber. To enlarge the measurement range limited by the free spectral range, a long short-term memory (LSTM) neural network was proposed. The original spectra were recorded by a low-cost spectrometer, and the intensity of scaled spectra was used to construct one-dimension (1D) spectral data. The results show that the coefficient of determination (R2) for the pressure prediction can reach 0.9996148 with the root mean square error (RMSE) equal to 2.559 × 10−2 MPa. Moreover, two-dimension (2D) data were obtained with the ascension algorithm of the Gramian angle field (GAF). A better R2 of 0.9999908 and a lower RMSE of 4.365×10−3 MPa can be obtained since the ascension algorithm could retrieve deeper features among the spectral data. The proposed approach can be adopted for a similar sensor structure, showing great potential in sensing applications requiring low cost and robustness.
PDF Article
More Like This
Cited By
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
Contact your librarian or system administrator
or
Login to access Optica Member Subscription