基于现代谱估计的光纤周界安防系统模式识别

Pattern recognition for fiber vibration system based on modern power spectral estimation

针对分布式光纤周界安防系统,为了有效识别事件类型并兼顾系统实时性,提出一种两级振动模式识别方案。第一级在时域上提取短时能量和短时过阈值率两个特征,与动态阈值进行比较,筛选疑似扰动样本;第二级对疑似扰动样本进行现代功率谱估计,提取各频段功率分布特征,并联合时域特征构建特征向量输入支持向量机(SVM)进行模式识别。

For distributed optical fiber perimeter security system, in order to effectively identify the event type and to balance the real-time, this paper proposes two-level vibration pattern recognition method. The first step is to judge initially the event type with short-time energy and short-time threshold crossing rate on time-do- main, and to screen the suspected intrusion signal samples by dynamic threshold comparison; the second step is to extract the power distribution features on frequency-domain through modern power spectral estimation on the suspected intrusion signal samples, and to combine with the time-domain characteristics as feature vec- tor through Support Vector Machine（SVM） to achieve pattern recognition.