Research On Intrusion Signal Analysis And Recognition Technology Of Optical Fiber Vibration Sensor Based On Machine Learning

With the development of social economy and the improvement of people’s security awareness,fiber optic perimeter security systems have been widely used and rapidly developed in the security field.Although the recognition accuracy of these security systems for intrusion events has been greatly improved,the amount of parameter data for the signal network model recognition algorithm is too large and requires a large amount of computational resources to be consumed.Therefore,this paper proposes a fibre-optic perimeter security system based on MobileNet,a lightweight convolutional neural network.The system greatly reduces the number of parameters of the network model while maintaining high recognition accuracy,thus saving computational resources.The main work of this paper is as follows:1.the basic principle of optical fibre sensing and the working principle of three common interferometric fibre sensors Sagnac type,Mach-Zander type and Michelson type are analysed,and Michelson interferometric fibre sensors are selected according to the engineering needs.2.For the collected fiber optic vibration signals,a pre-processing operation including endpoint detection and normalization was firstly applied;then the fiber optic vibration signals were characterized in the time domain,frequency domain and time-frequency domain.The short-time average energy and shorttime average amplitude are analysed in the time domain;the spectrogram is analysed in the frequency domain;finally,the time-frequency domain is analysed,and common time-frequency analysis methods are introduced,including the short-time Fourier transform,wavelet transform and HilbertHuang transform.The best time-frequency analysis method was determined by comparing the data collected in the laboratory,and the time-frequency map of the signal was extracted and used as input to the network model.3.Based on the initial MobileNet network model,a new MobileNet-based algorithm for fibre-optic perimeter security systems is proposed and experimentally evaluated to demonstrate that the algorithm is less complex while ensuring high accuracy.