Pattern Recognition Method of Fiber Optic Sensing Disturbance Signal Based on BILSTM

Abstract

The invention relates to a BiLSTM-based composite feature optical fiber sensing disturbance signal pattern recognition method, comprising the following steps: collecting different modes of vibration signals including optical fiber sensing disturbance signals and storing data, and adding type labels; time domain Feature extraction unit: Calculate the short-term energy and short-term over-level rate of the collected vibration signal, set the short-term energy and short-term over-level rate threshold, and preliminarily determine the intrusion disturbance signal according to the discrimination conditions; frequency domain feature extraction unit: Perform 4-layer wavelet packet decomposition on each vibration signal, solve the energy spectrum distribution of 16 sub-bands, concatenate short-term energy and short-term over-level rate to form a composite eigenvector, and normalize it. The normalized eigenvector is used as Input the feature vector; build a bidirectional LSTM network model, use the normalized feature vector as the input, and the event label as the classification output result, use the test sample to train the classifier, and realize the pattern recognition of the fiber sensor disturbance signal.

Description

technical field [0001] The invention relates to the field of optical fiber sensing signal identification, in particular to a method for classifying and identifying events of optical fiber sensing disturbance signals based on the BiLSTM algorithm. Background technique [0002] The distributed optical fiber sensing system has the advantages of high sensitivity, simple structure, long detection distance, less susceptible to interference, and no power supply. It has been widely used in peripheral security, oil and gas pipeline detection and other fields. In the field of perimeter security, the optical fiber sensing system uses optical fiber as the sensing element, which can be wound on the fence or buried in the ground. By collecting the sensing signal, the signal processing and analysis module is used to classify and identify the disturbance signal. The purpose of identifying the type of vibration intrusion signal is achieved. At present, the typical optical fiber perimeter pr...

AI Technical Summary

Problems solved by technology

The traditional methods of signal recognition mainly include short-term average zero-crossing rate ZC based on time-domain characteristics, differential short-term average zero-crossing DZC, short-term average energy E, differential short-term average energy DE, and disturbance duration. Analysis of time-amplitude information can characterize some local features of the signal, but cannot extract useful information such as waveform transformation; there is a wavelet packet decomposition method based on frequency domain feature analysis, which decomposes the signal into sub-frequency bands to extract energy features, extract features Vector can solve the problem of non-stationary signals; the method based on empirical mode decomposition EMD has insufficient theoretical basis and requires multiple iterative calculations, which is not efficient

Images