In the optical fiber perimeter security system, both deliberate invasion and environmental noise can cause optical fiber sensor vibration. It is very important to distinguish the invasion and no invasion events under ensuring the high sensitivity of system. In order to identify the various fiber optic vibration signals effectively, we propose two-stage discriminative method based on the invasion and environmental noise vibration signals short features on time-domain as well as scale energy distribution on complex wavelet domain to recognize the fiber optic signal. The first level use the time-domain characteristics of short-time energy and short-time zero crossing rate to judge whether there is a vibration happening. The second level use complex wavelet extraction the energy distribution features, combining with the time domain characteristics form the characteristic vector, Support Vector Machine (SVM) as the classifier to identify whether it is a intrusion signal and intrusion type. The experiment results show that this method can identify the intrusion signals and environment noise signals effectively, improve the system alarm rate and reduce the nuisance alarm rate.