Abstract: Optical fiber vibration sensor is widely used in the new generation of security monitoring system. The feature extraction and recognition methods of optical fiber vibration signal have become a research hotspot in the field of pattern recognition. The feature extraction and recognition methods of various optical fiber signals are summarized. These feature extraction methods decompose optical fiber vibration signals from the perspective of time domain, so different attribute characteristics of signals can be extracted. The empirical thresholds, neural networks and support vector machines are used to identify optical fiber vibration signals. Up to now, there is still a problem that the correct recognition rate of optical fiber intrusion events is not high. Vibration signal data of five kinds of optical fiber vibration signals, such as excavator mining, artificial digging, vehicle walking, personnel walking and noise, are visually analyzed. An effective method for feature selection of optical fiber vibration signal is proposed. According to the importance of optical fiber vibration intrusion events, identification tasks are completed in four stages, and the two-class task decision tree model and the constrained extreme learning machine algorithm are used to identify the type of intrusion events,which improves the correct recognition rate of all kinds of events.

Key words: intrusion event, experimental sample, constrained difference extreme learning machine (CDELM), constrained mixed extreme learning machine (CMELM), recognition rate