The phase-sensitive optical time domain reflectometer(Φ-OTDR) system uses the amplitude differential accumulation method to count the position of the vibration signal. There exists the problem of low recognition rate and need to be modified the N for different frequencies. An algorithm combining empirical mode decomposition(EMD) and neural network is proposed. The EMD is used to denoise and analyze the time domain signal at the vibration point, the signal characteristics are extracted by EMD and input to an improved neural network as features for training. The trained neural network is then employed to identify different vibration signals at different frequencies. The experiment shows that the recognition rate of vibration signal is 96.49%.