In recent years, pattern recognition technologies for distributed optical fiber vibration sensing have attracted more and more attention, aiming to intelligently recognize vibration events along with the optical fiber. Firstly, distributed optical fiber sensors for vibration detection are introduced. Secondly, this paper presents the state of the art of pattern recognition models used in distributed optical fiber vibration sensing systems. The feature extraction method, the model structure, and the processing performance are reported. As the results of the comparison, the support vector machine is a small sample learning method with a solid theoretical foundation and it has excellent generalization ability. The artificial neural network is suitable for massive data learning and multi-classification problems. Also, deep learning can learn more features information by a deep nonlinear network structure in an automated way, and thus has better accuracy and robustness. Furthermore, different applications of pattern recognition for distributed optical fiber vibration sensing are provided, including perimeter security, pipeline monitoring, and railway safety monitoring. Finally, the prospects of pattern recognition for distributed optical fiber vibration sensing are discussed.

Distributed optical fiber vibration sensor (DOFVS) is suitable for remote monitoring of vibration signals generated by external perturbations. In order to improve the sensing performance, various pulse coding techniques are applied in DOFVS, which have attracted an increasing amount of research attention. In this paper, recent advances of pulse coding methods used in DOFVS are reviewed, from pulse compression, including linear frequency modulation (LFM) and nonlinear frequency modulation (NLFM), pulse coding modulation based on the linear combination and correlation operation, such as Simplex code, Golay code and random code, as well as pulse sequence modulation (PSM), such as phase-shift pulses, differential frequency pulses, etc. Characteristics, working principles and the latest research results of pulse coding methods are discussed. Furthermore, some application progresses of pulse coding for performance enhancement in DOFVS are also introduced. By combining these technologies, the sensing performance could be improved in aspects of sensing distance, spatial resolution (SR), and frequency response, which show considerable potential applications in perimeter security, oil and gas monitoring, pipeline leakage, and structural health monitoring.

Phase-drift elimination is crucial to vibration recovery in the coherent detection phase-sensitive optical time domain reflectometry system. The phase drift drives the whole phase signal fluctuation as a baseline, and its negative effect is obvious when the detection time is long. In this paper, empirical mode decomposition (EMD) is presented to extract and eliminate the phase drift adaptively. It decomposes the signal by utilizing the characteristic time scale of the data, and the baseline is eventually obtained. It is validated by theory and experiment that the phase drift deteriorates seriously when the length of the vibration region increases. In an experiment, the phase drift was eliminated under the conditions of different vibration frequencies of 1 Hz, 5 Hz, and 10 Hz. The phase drift was also eliminated with different vibration intensities. Furthermore, the linear relationship between phase and vibration intensity is demonstrated with a correlation coefficient of 99.99%. The vibrations at 0.5 Hz and 0.3 Hz were detected with signal-to-noise ratios (SNRs) of 55.58 dB and 64.44 dB. With this method, when the vibration frequency is at the level of Hz or sub-Hz, the phase drift can be eliminated. This contributes to the detection and recovery of low-frequency perturbation events in practical applications.

A disturbance detection method with deep learning is proposed and experimentally demonstrated for OTDR. Compared with typical methods, better performance can be intensified by entirely using the multi-dimensional features of sensing data. The adaptability and anti-noise capacity are greatly advanced with the proposed method. In experiments, an ultra-high SNR of 53.98dB can be obtained in the best case. A spatial resolution of 1.06m and SNR of 38.49dB are achieved simultaneously when the pulse width is modulated to the narrowest. Moreover, additive noise and multiplicative noise with a level of 0.015 were applied, respectively. The SNR can reach 35.22dB and 42.39dB. With this approach, the stable high SNR is unprecedentedly improved. This method provides the potential for temporal-spatial disturbance detection in OTDR with strong background noise and extreme conditions.

The frequency division multiplexing (FDM) is an effective method to improve the detection ability of high-frequency disturbance in phase-sensitive optical time domain reflectometer (ϕ-OTDR). However, as the probe channels increase, crosstalk between channels will be generated. As a result, the recovery phase signal-to-noise ratio (SNR) will be deteriorated, which will reduce the detection capability of ϕ-OTDR. Therefore, a novel method for suppressing crosstalk noise by compressed sensing under multi-frequency ϕ-OTDR is proposed in this paper. The band-limited characteristics of vibration are deeply analyzed, and the frequency domain is chosen as the sparse domain. Noise without sparsity is discarded in the compression process. Furthermore, the noise cannot be recovered in the reconstruction process which uses the critical information containing the global property of phase signal. Consequently, the influence of crosstalk noise is reduced effectively. In experiments, compared with the restored signals without the proposed method, the average SNRs of the reconstructed 12 kHz sinusoidal vibration are respectively enhanced by 18.57 dB, 40.47 dB, and 41.67 dB in the sensing system with 6, 7, and 8 channels. In addition, the 24 kHz triangular wave signal and the disturbance composed of several different frequencies with random amplitudes can be detected and reconstructed. The proposed crosstalk noise suppression method provides a potential choice for ϕ-OTDR performance enhancement.

Aiming to reduce the complexity of Φ-OTDR system introduced by traditional distributed optical amplification technology in the pursuit of long-distance sensing, a novel detection range enhancement method with semantic image segmentation based on deep learning is presented. In this study, the Rayleigh backscattered curves for detecting vibration signals are used to construct the temporal-spatial image. The encoder-decoder structure and atlas spatial pyramid pool model are adopted to capture the boundary of vibration area and multi-scale context information from the temporal-spatial image. The semantic extraction of vibration position is then realized by binary classification of image pixels. In experiments, this method significantly improves the signal-to-noise ratio of the vibration location results. The signal-to-noise ratio can reach 37.84 dB, 34.28 dB, 34.09 dB, and 32.17 dB at 4 km, 10 km, 20 km, and 40 km, respectively. With this approach, the vibration detection range is enhanced to about 80 km without adding any traditional distributed optical amplification structure. Furthermore, within this enhanced detection range, the vibration signals with different frequencies and different intensities can be located and multi-point vibration can also be verified experimentally. This method provides a novel means for the detection range enhancement of Φ-OTDR system.