Note: Gaussian mixture model for event recognition in optical time-domain reflectometry based sensing systems

Abstract

We propose a novel approach to the recognition of particular classes of non-conventional events in signals from phase-sensitive optical time-domain-reflectometry-based sensors. Our algorithmic solution has two main features: filtering aimed at the de-nosing of signals and a Gaussian mixture model to cluster them. We test the proposed algorithm using experimentally measured signals. The results show that two classes of events can be distinguished with the best-case recognition probability close to 0.9 at sufficient numbers of training samples.The problem of the development of novel sensor techniques plays a crucial role in science and technology. One of the most important classes of sensing systems is distributed sensors, which are of great importance for the remote control of extended objects [1][2][3][4][5][6][7][8]. Phase-sensitive coherent optical time-domain reflectometry (Φ-OTDR) is a basic technique that can provide sufficient sensitivity and resolution for these distributed sensing systems [4][5][6]. Standard OTDR techniques use light sources with coherence lengths, which are shorter than pulse lengths. This can yield a sum of backscattered intensities from each scattering center, which allows one, e.g., to control splices and breaks in fiber cables [8]. On the contrary, in Φ-OTDR-based sensing systems [9][10][11][12][13][14], the coherence length of lasers is longer than their pulse length. An event near the optical fiber generates an acoustic wave that affects the fiber by changing the phases of the backscattering centers. An analysis of such signals can reveal their impact on the sensor and monitor located near fiber objects [9][10][11][12][13][14]. A key stage in implementing Φ-OTDRbased sensors is the development of an algorithmic solution to reveal unusual vibrations in the background.The problem of the recognition of non-conventional activity (a target event) consists of two closely related subproblems. The first is related to the de-noising procedure, which allows the detection of an event in the background with high probability. The second and much more important subproblem is the development of a classification methods aimed at clustering detected target events into predetermined classes. Due to the complex structure of the signals in such sensors, this is challenging [15][16][17][18][19][20][21][22][23].In addition to guaranteeing high accuracy of recognition, post-processing algorithms for Φ-OTDR-based sensors should be able to operate rapidly without significant computational resources. In other cases, their application in real-time distributed fiber optic sensing systems is substantially limited. In vibration sensing systems based on Φ-OTDR-based, there is at present no sufficiently fast and versatile algorithmic solution for the recognition of events. A promising direction for the solution of this problem is using of a machine learning toolbox, in particular neural networks and pattern recognition techniques [20][21][22][23]. Recent results [21][22][23] have shown that recognition algorithm...

A Novel DAS Signal Recognition Method Based on Spatiotemporal Information Extraction With 1DCNNs-BiLSTM Network

“…Thus at the second stage (from 2015-2017), more researchers focus on exploring various feature extraction and proper identification methods. The feature includes the magnitude [7], level crossing rate [22], and periodic gait characteristics [23] in time domain, and the frequency energy distribution of its Fast Fourier Transform(FFT) spectra [23], and the morphological features in space-time domain [12], and the spectra obtained by short-time Fourier Transform(STFT) [25], Wavelet or Wavelet packet energy spectra [25], [27], and Mel-Frequency Ceptral Coefficient (MFCC) [28] in time-frequency domain and etc. And the classification models include artificial neural networks(ANN) [6], [25], [27], Gaussian mixture model (GMM) [25], [28], Support Vector Machine(SVM) [7] and Relevance Vector Machine (RVM) [12] and etc.…”

Influence of the Laser Frequency Drift in Phase-Sensitive Optical Time Domain Reflectometry

“…2. Also we note that such procedure is the basic step of the recognition algorithm for Φ-OTDR based sensors [22]. In our tests, we are primary interested in links between the rate R det of correctly detected events, frequency drift behaviour, and parameters of the light source.…”

Phase-sensitive optical time-domain reflectometry with pulse mode EDFA: Probe pulse preparation

“…The form of probe pulse plays a crucial role for recognition procedures [10,16,17]. One can see that the resulting reflectograms obtained by probing the fiber by distorted and perfectly rectangular pulses (for comparison of the distorted and perfectly rectangle pulses, see Fig.…”

“…An analysis of such signals can reveal their impact on the sensor and monitor located near fiber objects. An important issue towards to practical implementation of such systems is a development of event recognition algorithms [10][11][12][16][17][18].…”