摘要
针对分布式光纤入侵监测系统在室外复杂环境下误报率过高的问题,提出了一种基于时/频域综合特征提取的入侵事件识别方法。使用自适应幅值门限信号切分算法找出有效振动信号片段,在此基础上提取平均片段间隔特征。选取最大能量片段作为主要研究对象,提取片段长度和峰均比特征,并对其进行小波包分解,生成频域能量分布特征,组成时/频域复合特征向量,使用高性能的支持向量机多分类算法进行模式识别。实验结果表明:该方法对行人脚踩、自行车轧过、拍击围栏和剪切光缆这4种典型入侵事件的平均识别正确率达到了98.33%,相比于仅提取时域或频域特征方法的识别正确率均有显著提高。该方法对光路光功率变化不敏感,能有效提升系统的实用性。
To reduce the high false alarm rate of the distributed fiber intrusion monitoring system in outdoor complex environment, this study proposes and demonstrates an intrusion event discrimination method based on integrated time/frequency domain feature extraction. First, a vibration fragment segmentation algorithm based on a self-adaptive amplitude threshold is developed to distinguish the vibrating part. On this basis, the average fragment interval feature is extracted. Next, the vibration fragment with the maximum energy is chosen as the research target, and the length and peak-to-average ratio are extracted in the time domain, whose energy distribution in the frequency domain is calculated according to wavelet packet decomposition and an integrated time/frequency domain feature vector is formed. Finally, one-versus-one support vector machine is used to classify four common intrusion events: footsteps of a passerby, bicycle rolling, knocking on the fence, and cutting of an optical cable. The experimental results show that the proposed method recognizes the abovementioned four common intrusion events with an average accuracy of 98.33%, which is much more accurate than the methods that only extract the time or frequency domain features. Moreover, the proposed method is immune to the optical power variation in light path. Thus, the proposed method is helpful to improve the utility of the system.
作者
彭宽
冯诚
王森懋
艾凡
李豪
刘德明
孙琪真
Peng Kuan;Feng Cheng;Wang Senmao;Ai Fan;Li Hao;Liu Deming;Sun Qizhen(School of Optical and Electronic Information,Huazhong University of Science and Technology,Wuhan,Hubei 4S0074,China;Wuhan Fisilink Microelectronics Technology Company Limited,Wuhan,Hubei 430074,China)
出处
《光学学报》
EI
CAS
CSCD
北大核心
2019年第6期338-348,共11页
Acta Optica Sinica
基金
国家自然科学基金面上项目(61775072)
湖北省自然科学基金创新群体项目(2018CFA004)
关键词
光纤光学
分布式光纤入侵监测系统
支持向量机
特征提取
模式识别
fiber optics
distributed optical fiber intrusion sensing system
support vector machine
feature extraction
pattern recognition
