深度学习下光纤围栏入侵告警方法  

Intrusion warning method of optical fiber fence based on deep learning

作　　者：乔文增 王兵[2] 张强[2] QIAO Wenzeng;WANG Bing;ZHANG Qiang(Center for Modern Information and Educational Technology,Shanghai Ocean University,Shanghai 201306,China;Shanghai Science and Technology University,Shanghai 201306,China)

机构地区：[1]上海海洋大学现代信息与教育技术中心,上海201306 [2]上海理工大学,上海201306

出　　处：《激光杂志》2020年第10期162-166,共5页Laser Journal

基　　金：国家级自然科学基金(No.5210523);上海理工大学2019年教师教学发展研究项目(No.CFTD194076)。

摘　　要：针对传统安防系统中入侵信号识别速度缓慢、误报率较高等问题,提出一种深度学习下光纤围栏入侵告警方法。首先利用稀疏自编码器对信号的数据特征逐层提取,借此获取信号中的有效信息,同时将不同类别的信号进行正确归类;其次建立光纤围栏入侵告警系统,利用端点测量判断是否存在扰动信号,并将采集到的数据作为训练集进行训练,实现扰动信号位置的精准定位;最后运用相空间重构对入侵告警信号深层次识别,将相空间重构加入维数形成复小波包,进行转换数据,输入小波包维数长度并生成入侵信号识别的特征集,利用主成分分析对原本特征集降维,减少光纤围栏系统误报率。仿真结果表明,所提方法可显著提升入侵信号识别精度,且识别效率较高,具有较好的鲁棒性。Aiming at the problems of slow recognition speed and high false alarm rate of intrusion signal in traditional security system,an intrusion warning method of optical fiber fence based on deep learning is proposed.Firstly,the sparse self encoder is used to extract data characteristics of signal layer by layer,so as to obtain effective information in the signal,and classify different types of signals correctly.Secondly,the fiber fence intrusion warning system is established,and the endpoint measurement is used to determine whether there is a disturbance signal,and the collected data is used as training set for training,so as to realize accurate location of disturbance signal.Finally,the phase space reconstruction is used to identify intrusion warning signal in depth,the phase space reconstruction is added to the dimension to form complex wavelet packet,the data is transformed,the dimension length of wavelet packet is input and the feature set of intrusion signal recognition is generated,the dimension of original feature set is reduced by principal component analysis,and the false alarm rate of optical fiber fence system is reduced.Simulation results show that this method can significantly improve the accuracy of intrusion signal recognition,has high recognition efficiency and good robustness.

关 键 词：深度学习模型 光纤围栏 入侵识别 相空间重构 

分 类 号：TN974[电子电信—信号与信息处理]