分布定位式光纤振动传感系统的事件识别

Event recognition based on the distributed located fiber sensor

本文提出了一种对分布式光纤声传感器的入侵事件分类方法。该方式采用小波包去噪方式对原始信号进行去噪;将去噪后的原始信号进行小波变换,得到原始信号的小波时频图;构建双输入型的卷积神经网络,将滤波后的原始一维信号直接输入到一个三层的1-D CNN中、滤波后得到的二维小波时频图直接输入到一个两层的2-D CNN中;将两种CNN输出的特征输入到支持向量机(SVM),使用SVM对事件进行分类。本文中主要识别3种振动事件:汽车通过、挖掘机挖掘和破路机工作。实验结果表明,所提方式对实际环境中3种振动事件的识别准确率平均可以达到96%,并且识别时间仅为0.61s。

In this paper,a novel approach to classify intrusion events from distributed fiber optic acoustic sensors is proposed. The approach consists of four main steps. At the first step,the original signal is denoised using wavelet packet denoising. At the second step,the denoised original signal is transformed with wavelet to obtain the wavelet time-frequency map of the original signal. At the third step,a two-input convolutional neural network is constructed. The filtered original 1-D signal is directly fed into a three-layer 1-D CNN,and the filtered 2-D wavelet time-frequency map is directly fed into a two-layer 2-D CNN. At the fourth step,the features output from the two CNNs are fed to a support vector machine(SVM),which is used to classify the events. A phase-sensitive optical time-domain reflectometer(φ-OTDR)sensing system is also used. In this paper three main vibration events are identified:car passing,excavator digging,and road breaker working. The experimental results show that the recognition accuracy of the three vibration events in the real environment using the proposed approach can reach 96% on average,and the recognition time is 0.61 s.