At present, the recognition of vibration events by distributed fiber optic vibration sensing is easy to overfit and the generalization ability is insufficient. We propose a distributed fiber optic vibration recognition method based on fractional Fourier transform (FrFT). The method uses FrFT for the time-frequency conversion of the time-domain vibration signals, which means adding a new analysis dimension compared with the traditional time-domain method. Compared with the traditional time-frequency domain method, this method not only solves the problem of mutual constraints of time resolution and frequency resolution but also enhances the feature richness of time-frequency data. It is more conducive to the learning of deep models and can effectively prevent model overfitting. Outdoor field experiments show that the recognition accuracy of the FrFT method reaches 98.5%, and the accuracy can still maintain more than 98% on the special generalization test set. At the same time, a more reliable evaluation index f1-score is introduced. The f1-score is the harmonic mean of precision and recall. It is used to comprehensively evaluate precision and recall. The f1-score of each event recognized by this method is higher than 0.975.