Distributed optical fiber sensing technology has been widely used in the safety monitoring of pipeline transportation. Accurately extracting and classifying the characteristics of different optical fiber vibration signals is a research hotspot in recent years. Aiming at the defects that the traditional time-frequency analysis method needs to manually set the basis function and cannot eliminate the interference of high-frequency noise when studying the optical fiber vibration signal, the local mean decomposition (LMD) method with adaptive characteristics is used to process the signal, and a new method based on LMD is proposed in this paper. Decomposed feature extraction and recognition methods. First, LMD is performed on the original signal to obtain several product function components; then, the signal is reconstructed by the principle of autocorrelation, and the sample entropy features and energy features of the reconstructed signal are extracted; finally, the above features are fused and sent to the support vector machine for training recognition. The experimental results show that the method can effectively identify different vibration types, and has a high accuracy rate.