A method for the recognition of fiber-optic perimeter vibration signals is proposed based on local mean decomposition (LMD) and serial feature fusion (SFF), in which the effect of noise is first suppressed to extract the relevant information of vibration signals, then the SFF is conducted to get the feature vectors with the ability of accurate description, and finally the probabilistic neural network (PNN) algorithm is adopted for learning and classification. The proposed method is validated by different single-vibration signals and vibration signals under the stormy weather interference. The results show that, by the proposed method in the above two cases, the average correct-recognition rates reach 96.0% and 96.7%, and the recognition time is 0.87 s and 0.91 s, respectively. The proposed method is superior to the traditional LMD algorithm and the SFF-PNN algorithm in the sensitive information recognition and feature extraction.