The invention belongs to the field of pipeline monitoring, and particularly relates to a pipeline safety monitoring system based on deep learning for solving the problem that the existing interferencetype optical fiber early warning system has high missing report and false report rates and cannot stably identify an intrusion event category. The pipeline safety monitoring system provided by the invention comprises an optical fiber detection module for acquiring a detection signal and a reference signal; a signal processing module for acquiring time domain and frequency domain statistical features of an interference signal formed by the detection signal and the reference signal; and an event classification module for judging an even feature category by using an event classification model constructed based on a deep neural network to obtain an event category. The pipeline safety monitoring system provided by the invention improves the recognition accuracy of events, reduces the missing report and false report rates, and timely and accurately discovers intrusion events.