Extraction of double-peak Brillouin frequency shift based on ANN for nonuniform strain within BOTDR spatial resolution

For practical engineering applications, Brillouin optical time domain reflectometry (BOTDR)'s double-peak Brillouin gain spectrum (BGS) can be measured when the strain is not uniform within spatial resolution. This double-peak BGS will affect the extraction accuracy of Brillouin frequency shift (BFS) and the following disaster monitoring. A detection method based on artificial neural network (ANN) is proposed for the double-peak BFS extraction. The ANN model is trained to deal with multi-parameters cases, including different frequency ranges, signal-to-noise-ratios, spectral widths, etc. The retraining of ANN model for different single-mode optical-fibers is not necessary. After training, the ANN model successfully detects the double-peak BFS from both the simulated and experimental data. The ANN model based BOTDR system is applied to the urban safety monitoring of underground pipe gallery. Under 5 m spatial resolution, 20 cm strained fiber can be detected, and the standard deviation of BFS can be as low as 1 MHz in experiment.