High-Accuracy Event Classification of Distributed Optical Fiber Vibration Sensing Based on Time–Space Analysis

Abstract

Distributed optical fiber vibration sensing (DVS) can measure vibration information along with an optical fiber. Accurate classification of vibration events is a key issue in practical applications of DVS. In this paper, we propose a convolutional neural network (CNN) to analyze DVS data and achieve high-accuracy event recognition fully. We conducted experiments outdoors and collected more than 10,000 sets of vibration data. Through training, the CNN acquired the features of the raw DVS data and achieved the accurate classification of multiple vibration events. The recognition accuracy reached 99.9% based on the time–space data, a higher than used time-domain, frequency–domain, and time–frequency domain data. Moreover, considering that the performance of the DVS and the testing environment would change over time, we experimented again after one week to verify the method’s generalization performance. The classification accuracy using the previously trained CNN is 99.2%, which is of great value in practical applications.

Intrusion signal recognition method based on φ-OTDR fiber distributed sensing research progress

“…9, and then the valid IMF components are judged using Pearson correlation coefficient (PCC), and finally the signals are identified using one-dimensional convolution, and the identification accuracy of the intrusion signals collected in the experimental environment can reach 98.3%. In 2022, Ge 53 et al normalized the data in order to obtain the features of the original DVS only, and achieved an accurate classification of multiple vibration events by identifying the spatio-temporal data, and the accuracy could reach 99.9%. In the same year, Xu 54 et al proposed to suppress the random noise of the original data using moving average method, record the difference between two adjacent Rayleigh backward scattering record channels, and normalize the difference results, and convert the processed 2D data into images for network training and testing.…”

A machine learning algorithm for event recognition and classification in -OTDR system based on lenet neural network

“…The working principle of Ф-OTDR technology is that when the external disturbance action is used for sensing the optical fiber, the refractive index of the sensing fiber will be changed [5,6], so that the Rayleigh scattered light will produce phase modulation, and the intensity or phase information of the backward Rayleigh scattered light pulse signal in the fiber can be distributed sensing. After the coherent pulsed light is injected into the sensing fiber through the circulator, the generated backward Rayleigh scattered light is returned to the front end of the fiber, and is received by the photodetector through the circulator, and the external strain change information is restored by demodulation by the demodulation unit [7].…”

Few-Shot Classification with Meta-Learning for Urban Infrastructure Monitoring Using Distributed Acoustic Sensing

“…One notable approach in [ 11 ] introduced a self-supervised DL method aimed at improving DAS measurements via mitigating spatially incoherent noise with unknown characteristics. In [ 12 ], different pre-processing methods for the input to various DL models were compared, and an accuracy of 99.2% in a four-way classification task was achieved, highlighting the importance of the initial data representation fed into a DL model. In [ 13 ], classical ML methods and DL methods were compared, and it was concluded that the best approach depends on the data regime: In the low-data setting (usually at a project’s beginning), classical ML approaches dominate DL methods, as neural networks tend to overfit on spurious correlations.…”