Based on the application of distributed optical fiber sensing system in the field of perimeter security monitoring, there are problems such as slow response speed and low recognition rate. Although the recognition rate of the traditional convolutional neural network is very high, its huge amount of parameters makes industrial deployment difficult and the recognition response speed is slow. This paper introduces the lightweight convolutional neural network MobileNet, which uses depth-separable convolution to replace the traditional convolution, which greatly reduces the amount of model parameters. This paper uses MobileNet as the benchmark network to implement a one-dimensional lightweight network based on MobileNet-18 Φ-OTDR perimeter intrusion event recognition, compared the network recognition rate and recognition speed under different structures through experiments, and selected MobileNet-18 as the best model under the condition that the accuracy of the model would not be greatly reduced. In the experiment, six perimeter fiber intrusion signals of climbing, cutting, wind blowing, lifting, pulling and walking were collected. Among the six types of fiber intrusion signal recognition, MobileNet-18 achieved a recognition rate of 98.33% and a response time of 9.27 ms.