一种分布式光纤传感系统的信号识别方法

A Signal Recognition Method for Distributed Optical Fiber Sensor System

分布式光纤传感系统利用光纤作为传感元件和和传输介质,对环境中的物理量进行测量.以往的算法都是计算机直接将采集到信号进行定位计算并发出警报,由于该系统具有极高的灵敏度,在实际应用中接收到大部分信号都是无用的自然界声音,导致程序运行时间长,内存占用率高,限制了系统的效率.本文在阐述分布式光纤传感器原理的基础上,提出了一种基于支撑向量机(SVM)对分布式光纤传感系统的信号进行模式识别的方法,SVM算法在解决小样本、非线性及高维模式识别中有许多特有的优势.首先提取出光纤传感信号的Mel频率倒谱系数(MFCC),然后以MFCC特征训练SVM模型,达到模式识别的目的.实验结果表明,该方法能够准确识别剪切、敲击等入侵信号和吹风、下雨等非入侵信号,可以有效排除自然界中风雨等的干扰,具有很强的实用性.

The distributed optical fiber sensor system measures ambient signals with the application of optical fiber as both sensing element and transfer element.In the past,the computer received signals are directly processed to detect position and send out alarm.The received signals are useless natural sounds in many cases because of high system sensitivity.Based on describing the principle of distributed optical fiber system,this article proposes a signal recognition method based on supporting vector machine（SVM）for distributed optical fiber sensor system.SVM algorithm has unique advantages in solving samples-limited,non-linear and high-dimensional pattern recognition problem.Firstly,Mel frequency cepstrum coefficient（MFCC）of the signal is extracted;secondly,SVM model is trained by MFCC for pattern recognition.The experiment result indicates that this method had high accuracy in the recognition of intrusion signals like cutting and striking,and non-intrusion signals like blowing and raining.The proposed method eliminates the influence of wind and rain and has high practical application value.