To reduce the time-consuming and misinformation of one dimensional signal recognition by the pre-warning system in a Phase-sensitivity Optical Time-domain Reflectometer(Φ-OTDR), a new method to acquire two dimension signals by the Φ-OTDR pre-warning system and to recognize events based on Relative Vector Machine(RVM) classifier was proposed. Firstly, the spatial and temporal two dimension signal was taken as an image and the image processing method was used for the threshold segmentation of different events according to the image characteristics. Then, the proposed feature extraction method based on morphology was used to extract different signal features by using the amplitude, area, shape and internal of region as feature vectors. Finally, the RVM classifiers and "one to one" strategy were used for multi-class recognition. The experiments on three pipeline safety events show that the feature extraction method proposed in this paper greatly improves the recognition accuracy with less computation time, the accuracy has been reached to 97.8% and the computing time is less than 1 s. As compared with traditional methods, the algorithm has better performance, thus is very suitable for the pre-warning system online monitoring of Φ-OTDRs.