基于SVM算法的分布式光纤扰动传感系统快速检偏技术研究  

Research on fast polarization detection technology of distributed optical fiber vibration sensing system using SVM algorithm

作　　者：徐中原 刘琨 江俊峰 孙振世 马鹏飞 郭海若 周子纯 刘铁根 XU Zhong-yuan;LIU Kun;JIANG Jun-feng;SUN Zhen-shi;MA Peng-fei;GUO hai-ruo;ZHOU zi-chun;LIU Tie-gen(Key Laboratory of Optoelectronic Information Technology,Ministry of Education,Institute of Optical Fiber Sensing,School of Precision Instruments and Optoelectronics Engineering,Tianjin University,Tianjin 300072,China)

机构地区：[1]天津大学精密仪器与光电子工程学院光电信息技术教育部重点实验室,天津大学光纤传感研究所,天津300072

出　　处：《光电子．激光》2020年第10期1101-1109,共9页Journal of Optoelectronics·Laser

基　　金：国家重大科学仪器设备开发专项(2013YQ030915)资助项目。

摘　　要：针对基于双马赫-曾德尔干涉的分布式光纤扰动传感系统对传感光信号的偏振状态需要进行实时监测以便及时进行偏振控制的问题,本文提出一种基于支持向量机(SVM)的偏振状态快速判别算法。该方法对采集到的两路干涉信号进行做差处理,得到原始数据。提取原始数据中的过零率与峰值的绝对和作为分类器的输入特征向量,使用Verilog HDL语言编写支持向量机(SVM)分类识别算法,借助现场可编程门阵列(FPGA)的硬件并行结构和流水线技术实现分类器权系数的快速迭代求解。通过提取未知偏振状态信号的特征向量,并将其输入到训练好的支持向量机模型中可以实现高效率的偏振状态识别。实验结果表明:本方法可以快速并准确的实现对系统偏振状态的识别判断,平均识别率达到93.25%,分类模型训练时间在100 ms以内,平均识别响应时间在8 ms以内。Aiming at the problem that the distributed optical fiber vibration sensing system based on dual Mach-Zehnder interferometer needs to monitor the polarization state of the sensing optical signal in real time in order to carry out polarization control in time,an algorithm based on support vector machine(SVM)to quickly judge the polarization state has been proposed in this work.This method makes a difference between the two collected interference signals to obtain the original data.Extract the zero-crossing rate and the absolute sum of peak value in the original data to construct the input feature vector of the classifier,use Verilog HDL language to write SVM classification recognition algorithm,with the help of field-programmable gate array(FPGA)hardware parallel structure and Pipeline technology realize fast iterative solution of classifier weight coefficients.By extracting the feature vector of the unknown polarization state signal and inputting it into the trained support vector machine model,high-efficiency polarization state recognition can be achieved.The experimental results show that this method can quickly and accurately judge the polarization state of the system.The average recognition rate of 93.25%is achieved,the training time of the classification model can be controlled less than 100 ms,and the mean recognition response time is within 8 ms.

关 键 词：光纤传感 现场可编程门阵列 支持向量机 过零率 峰值的绝对和 快速检偏 

分 类 号：TN29[电子电信—物理电子学]