A Positioning and Recognition Method Based on Bilinear Time–Frequency Transform and CNN for Long-Range Asymmetric Interferometer Vibration Sensor

High-precision intrusion positioning and recognition are crucial to promoting the practical application of long-distance distributed optical fiber vibration sensing systems in the engineering field. In this article, we propose a method for positioning and recognition in a long-distance asymmetric fiber-optic vibration sensing system based on bilinear time-frequency transform (BTFT) and convolutional neural network (CNN). First, a mean filter is conducted to obtain the endpoints of the vibration signal (VS). Second, the time-frequency feature of the VS is obtained by an affine smoothed pseudo Wigner-Ville distribution (ASPWVD), avoiding the incorrect positioning results caused by the phase mismatch of two signals. Third, the vibration position can be acquired by the time delay value obtained from the cross correlation. Meanwhile, a model capable of recognizing multiple types of perturbation events (PEs) is constructed by training the dataset containing ASPWVD of multiple VSs in CNN. Experimental results demonstrate that 94.8% of the positioning error reaches ±20 m, and the average positioning time is 302 ms, implementing the real-time positioning. The average recognition success rate of the five PEs and the average test time reaches 96.5% and 143 ms, respectively. Our work carries out the pattern recognition of PEs in the process of positioning. It is demonstrated that our work shows great prospects in the field of perimeter security.