A fast wavelength detection method based on OTDR and 1-DDCNN in series overlapping spectra

Abstract

To improve the rapidity and reliability of the subway tunnel fire monitoring system and realize the monitoring of small fire sources, the method of regional subsection demodulation is used to group several adjacent nearly identical ultra-weak fiber Bragg gratings (UWFBGs) sensors into a group, and the 1-D dilated convolutional neural network (1-DDCNN) is used to demodulate the same group of highly overlapping sensing signals. The well-trained 1-DDCNN model can achieve extremely low signal demodulation error. The experiment shows that the UWFBG demodulation scheme proposed in this paper improves the detection accuracy of the subway tunnel fire monitoring system and shortens the detection time. The root-mean-square error (RMSE) of four highly overlapping peak wavelengths of sensing signals is less than 1.5 pm, and the average demodulation time is less than 30 ms.