采用单模-多模-单模(SMS)光纤结构的光路,针对施加在多模光纤上的入侵信号,提出了基于短时傅里叶变换(STFT)和卷积神经网络(CNN)相结合的模式识别分类方法。该方法对入侵信号进行STFT以获得时频图,制作成训练集和测试集;将训练集输入到三种网络模型中进行训练,根据工程应用指标选择合理的网络模型;利用网络模型对测试集进行分析,得到入侵信号的识别结果。采用4种入侵信号对该方法的有效性和实时性进行验证。结果表明,该方法可以高效识别人为入侵信号和非人为入侵信号,并可以通过增加含有不同类型噪声的入侵信号种类和数量来验证此方法的稳健性,减少了入侵信号的漏报率和误报率,提高了SMS光纤结构在周界防区模式识别中的应用价值。

A single mode-multimode-single mode (SMS) optical fiber structure is adopted, and a pattern recognition classification method is proposed based on the combination of short-time Fourier transform (STFT) and convolutional neural network (CNN) to deal with the intrusion signals which are applied on the multimode fiber. The proposed method initially performs STFT on the intrusion signal to obtain the time-frequency map and subsequently creates a training set and a test set. Further, the training set is input into three network models for training, and a reasonable network model is selected according to the engineering application index. Finally, the identification result of the intrusion signal is made to the test set through the network model; furthermore, the validity and real-time performance of the method are verified using four intrusion signals. The results denote that the proposed method can effectively identify artificial and non-human intrusion signals; in addition, the robustness of this method can be verified by increasing the types and quantities of intrusion signals with noises, thereby reducing the alarm failure and false alarm rate of the intrusion signals and improving the application value of the SMS fiber structure in perimeter defense area pattern recognition.