基于EMD分解与1-D CNN算法的光纤振动信号的识别

Recognition of optical fiber vibration signals based on VMD_CNN algorithm

为提高基于相敏光时域反射计(Φ-OTDR)的分布式光纤声传感系统(DAS)对入侵振动事件的识别准确率,提出一种基于经验模态分解(EMD)与一维卷积神经网络(1-D CNN)相结合的识别方式。该方式首先使用EMD将振动信号分解为m阶本征模函数(IMF),然后使用皮尔逊相关系数(PCC)判断出有效的IMF分量,将有效的IMF分量使用小波阈值去噪算法(WTD)进行去噪,对所有去噪后的IMF分量求和得到重构信号,最后使用1-D CNN对重构信号进行识别。实验证明该识别方式能快速完成对识别模型的训练,训练时间小于3 min,并且能有效识别在实际环境中采集的入侵振动信号,对入侵信号的识别准确率可达98.3%。

In order to improve the recognition accuracy of intrusive vibration events by a distributed fiber-optic acoustic sensing system(DAS)based on a phase-sensitive optical time-domain reflectometer(Φ-OTDR),a recognition approach based on empirical modal decomposition(EMD)combined with a one-dimensional convolutional neural network(1-D CNN)is proposed.The EMD is firstly used to decompose the vibration signal into m-order intrinsic mode function(IMF),then the Pearson correlation coefficient(PCC)is used to determine the effective IMF components,the effective IMF components are denoised using the wavelet threshold denoising algorithm(WTD),the reconstructed signal is obtained by summing up all the denoised IMF components,and finally the 1-D CNN is used to identify the reconstructed signal.It is proved that this identification method can quickly complete the training of the identification model,the training time is less than 3 min,and can effectively identify the intrusion vibration signals collected in the real environment,and the identification accuracy of the intrusion signals can reach 98.3%.