基于波网络的相位敏感光时域反射系统模式识别方法研究

针对相位敏感光时域反射系统中传统人工特征提取和模式识别方法实时性差、准确率低的问题,提出一种波网络模式识别方法。该方法通过因果空洞卷积结构充分分析光纤振动信号的时序因果性,通过残差块结构使模型更快收敛,以实现更高的识别准确率和效率。实验结果表明,在对手拍、脚踩、棒击三种信号识别时,与一维卷积神经网络结构和长短期记忆网络结构方法相比,该方法识别准确率高达99.85%;且训练耗时最少,低至96 s,测试耗时也仅为30 ms,满足应用实时性的要求。该模式识别方法既具有高准确率又具有高实时性,对于φ-OTDR系统在周界安防中的应用推广具有重要意义。

Phase-sensitive Optical Time-domain Reflectometric System Pattern Recognition Method Based on Wavenet

To improve on the poor real-time performance and low accuracy of traditional manual feature extraction and pattern recognition method in the Phase-Sensitive Optical Time-Domain Reflectometric System(φ-OTDR),a new pattern recognition method based on wavenet is proposed.This method fully analyzes the temporal causality of optical fiber vibration signals through the causal dilated convolutions,and makes the model converge faster by the residual block structure,so as to achieve higher recognition accuracy and efficiency.The experimental results show that,when three signals are recognized,namely,hand tapping,foot stepping,and stick striking,compared with the other two common methods,onedimensional convolutional neural network structure and long short-term memory structure,the recognition accuracy as high as 99.85%is achieved with the proposed method.And it consumes the least amount of training time,as short as 96 s.Also,its signal detection process takes only 30 ms,which can meet the realtime application's requirement.The proposed pattern recognition method has high accuracy and good realtime performance.It should be of great significance for the application and popularization of φ-OTDR systems in perimeter security and similar areas.