A new multi-area perimeter security system based on optical fiber Michelson interferometer and pattern recognition technique is proposed and demonstrated. The polarization-insensitive optical fiber Michelson interferometer is used as the vibration sensor for perimeter intrusion detection system (PIDS) which reduces the polarization-induced fading. Strain measurement of fiber bonding on host material is analyzed. Noise suppression, behavior analysis and alarm judgement are realized by using level crossing (LC) method and pattern recognition (PR) technique. The experiment results show that the system is able to analyze the original signal with LC method and extract the feature values with response time less than 0.5 s. Behavior pattern is generated through neural network training. Rain-induced noise and other environmental noise are eliminated by dynamic means algorithm (DMA). So the all-fiber perimeter security system with low lost, high reliability and high praclicability is realized.