ΦOTDR光纤入侵特征提取算法研究及实现

发布时间：2018-03-21 22:15

本文选题：光纤预警系统　切入点：相关系数　出处：《北方工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于 Φ-OTDR(Phase-Sensitive Optical Time-Domain Reflectometer,相位敏感光时域反射计)的分布式光纤预警系统是一种新型的安防系统。其具有抗干扰性强、灵敏度高等优点,可进行长距离、并发的入侵行为实时监测。然而,光纤预警系统实际应用中检测到的入侵信号类型繁多、特性复杂,需对入侵信号进行有效的预处理及特征提取。本文围绕这一技术难点,开展相应的算法及工程实现研究,主要完成以下内容:本文首先总结了光纤预警系统现状及已有入侵信号特征提取方法,归纳出己有方法及其实现瓶颈,引出本文的研究思路及内容。接着,介绍了本课题所构建的光纤预警系统组成及工作原理。在光纤入侵信号预处理方面,根据光的双折射现象对入侵信号进行时域相关性分析,发现时域上虚警与入侵信号的相关系数存在明显差异。为此,本文提出一种基于时域相关系数的预处理方法。其中,为提升互相关系数计算效率,针对其中制约算法效率的小波去噪和相关计算两个模块,进行了实现层的优化。实测数据验证了本文预处理实现方法能有效去除虚警信号。在光纤入侵信号特征提取方面,本文研究了多种信号提取方法及其工程实现。在时域上提取占空比和基音周期特征,频域上提取频率中心特征,时频域上提取小波系数能量占比特征;由多维特征构建的特征向量对机械与人工信号具有很强的区分性。其中,为有效提取基音周期特征,本文提出了基于端点检测预筛选的基音周期特征提取方法,有效提高了该特征提取的准确性。特征提取算法的工程实现是本文的重要研究内容。针对光纤入侵信号特征提取的特点,本文基于DSP平台设计特征提取实现架构,使用模块化思想进行功能划分,主要包括:DSP与上位机数据交互逻辑,DSP间的数据同步,DSP数据存储。利用现场实际数据对本文提出的光纤入侵信号预处理及特征提取方法及其实现,进行了验证。
[Abstract]:The distributed optical fiber early warning system based on 桅 -OTDR Phase-Sensitive Optical Time-Domain reflection (Phase sensitive Optical time Domain reflectometer) is a new type of security system. The real-time monitoring of concurrent intrusion behavior. However, in the practical application of optical fiber early warning system, there are many types of intrusion signals detected and the characteristics are complex, so it is necessary to carry out effective preprocessing and feature extraction of intrusion signals. This paper focuses on this technical difficulty. The main contents are as follows: firstly, this paper summarizes the current situation of optical fiber early warning system and the existing intrusion signal feature extraction methods, and concludes the existing methods and their implementation bottlenecks. This paper introduces the research ideas and contents of this paper. Then, the composition and working principle of the optical fiber early warning system are introduced. According to the birefringence of light, the time-domain correlation analysis of the intrusion signal is carried out in the aspect of optical fiber intrusion signal preprocessing. It is found that the correlation coefficients of false alarm and intrusion signal are obviously different in time domain. Therefore, a preprocessing method based on time domain correlation coefficient is proposed in this paper. Aiming at the two modules of wavelet denoising and correlation calculation, which restrict the efficiency of the algorithm, the realization layer is optimized. The experimental data verify that the proposed preprocessing method can effectively remove false alarm signals. In this paper, a variety of signal extraction methods and their engineering implementation are studied. The characteristics of duty cycle and pitch period are extracted in time domain, frequency center feature is extracted in frequency domain, and wavelet coefficient energy ratio feature is extracted in time and frequency domain. The feature vectors constructed from multi-dimension features have strong distinctions between mechanical and artificial signals. In order to extract pitch cycle features effectively, a method of extracting pitch periodic features based on endpoint detection pre-screening is proposed in this paper. The engineering implementation of feature extraction algorithm is an important research content in this paper. According to the feature extraction characteristics of optical fiber intrusion signal, this paper designs a feature extraction implementation framework based on DSP platform. This paper uses the modularization idea to divide the function, mainly including the data synchronization and DSP data storage between the component DSP and the host computer data interaction logic DSP. The method and realization of the fiber optic intrusion signal preprocessing and feature extraction proposed in this paper are analyzed by using the field actual data. Verification was carried out.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP277

【参考文献】