Research On Identification Of Distributed Sagnac Fiber Intrusion Signal Based On Variational Mode Decomposition

As the society's perimeter security requirements continue to increase in the standards of distance,recognition rate and cost.The traditional perimeter security system have been unable to meet people's needs.The distributed vibration optical fiber sensor adapts to the society's demand for intelligent,long-distance,low-cost and high-precision security systems through its own advantages.The optical fiber perimeter security system has antielectromagnetic interference and anti-corrosion properties,which could be flexibly buried in the ground,embedded in the wall.Without obvious external equipment,it has certain concealment.The optical fiber perimeter security system can monitor and detect the intrusion signals from a long distance.And there are advantages of it like high positioning accuracy for intrusion signals and strong tolerance to complex external environments.At present,distributed vibration optical fibers has been widely used in railway rails,cables,border security and other important fields.Optical fiber perimeter security system has the problem of high false alarm rate of various intrusion signals.How to reduce the false alarm rate of the system to make it better adapt to complex environments and improve practicality becomes particularly important.This paper is aiming at the problem that distributed Sagnac fiber has a poor recognition rate of multiple types of intrusion signals,the endpoint detection,feature extraction and intrusion recognition of intrusion signals are studied respectively.The Variational Mode Decomposition(VMD)algorithm is introduced for intrusion signal recognition research to extract more representative features and improve the recognition rate of signals.This paper analyzed the basic principles and characteristics of three typical types of sensors in the fiber perimeter security system: Mach-Zehnder(M-Z)type,Michelson type and Sagnac type,mainly analyzed the advantage of Sagnac type.Through the basic simulation of Empirical Mode Decomposition(EMD)and VMD algorithm,the signal processing effects of them are studied,and the selection of VMD parameters is analyzed.The experimental results show that VMD can better suppress the modal aliasing phenomenon rather than EMD.Secondly,aiming at the low recognition rate of intrusion signal,the whole process of vibration fiber intrusion signal recognition is analyzed.This paper focuses on the endpoint detection and feature extraction algorithms of four types of intrusion signals:tapping,flapping,walking,and throwing stones.Because of the weak characteristics of the intrusion signal caused by the environmental noise,and the collected vibration fiber signal is preprocessed.Introduce spectral entropy method is introduced to denoise the intrusion signal and improve the signal-to-noise ratio.In implementing the endpoint detection algorithm of intrusion signals,this paper proposes an endpoint detection algorithm based on the fusion of short-term logarithmic energy and spectral entropy.Compared with the endpoint detection effects of spectral entropy and short-term logarithmic energy,the fusion algorithm has a better effect on improving the accuracy of endpoint detection.Since the extracted features cannot characterize all kinds of intrusion signals well,VMD is introduced into the processing of intrusion signals.By performing VMD decomposition on the extracted intrusion signals.Features like spectral entropy,energy ratio and kurtosis has extracted from each Intrinsic Mode Function(IMF).And then extract the time-domain features of the intrusion signal to form a multi-dimensional feature vector.As for the extraction of multi-dimensional features,in order to reduce the feature dimension and improve the effectiveness of the feature,correlation analysis is used to simplify the feature dimension and improve the operation efficiency.Finally,the Support Vector Machine(SVM)and B-P algorithms are introduced respectively to classify the VMD,EMD and wavelet feature vectors of various intrusion signals.The experimental results illustrate that the fusion endpoint detection method based on short-time logarithmic energy and spectral entropy proposed in this paper improves the detection accuracy of the intrusion signal and enhances the system's antinoise ability.The time domain and frequency domain feature extraction of each mode function of intrusion signal is realized by using VMD decomposition.Through uncertainty analysis,the feature dimension is reduced,and the screening can characterize various intrusive signals better.The recognition and classification of intrusion signals are realized based on SVM,and the overall recognition rate of all kinds of intrusion signals reaches 98.0%.Compared with the recognition rate of wavelet 91.4% and EMD92.6%,this algorithm has higher recognition rate.