Research On Analysis Technology Of Detection Signal Of Optical Fiber Perimeter Security System Based On Pattern Recognition

In recent years,as the optical fiber perimeter security system is widely used in the security protection of large defense areas such as warehouses,oil and gas pipelines,how to improve the real-time and accuracy of the security system has become the focus of research in the field of security.The optical fiber perimeter security system uses fiber optic vibration sensor to collect fiber optic vibration signal,but the collected signals are chaotic and complex,so how to effectively process the signals and identify and classify them is an urgent problem to be solved.Based on the above research background,this paper conducts research on the signal processing and identification technology of the optical fiber perimeter security system.Relevant work and proposed innovations are as follows:(1)Explain the working principle of different interference type optical fiber vibration sensors,compare and analyze the advantages and disadvantages of each sensor and the applicable scenarios of each sensor,and explain the reasons for using Michelson fiber-optic vibration sensors.(2)Study the processing methods of optical fiber vibration signals,on the basis of feature extraction in time domain,frequency domain and wavelet domain,innovatively proposed to apply the double threshold endpoint detection method and Mel coefficient feature extraction method to the data processing section,and apply the above method to the processing of optical fiber vibration signal.(3)Mainly explains the reason for choosing the traditional KNN algorithm,introduce the basic principle and implementation process of traditional KNN algorithm,combined with the classification criteria of this thesis,analyze the reasons for choosing Euclidean distance among various distance measurement methods.(4)Based on the analysis of the traditional KNN algorithm,this paper innovatively proposes improvements to the KNN algorithm,mainly including the following two aspects:choose the best K value by the cross-validation method to get the highest recognition accuracy,apply KD tree to KNN algorithm to improve algorithm efficiency,and improve system recognition accuracy through distance weighting.Finally,comparing and analyzing the traditional KNN algorithm and improved KNN algorithm through experiments.