A recognition method for multiclass intrusion events based on zero-crossing rate feature extraction is proposed; in this approach, the intrusion signal is processed by segments, and the zero-crossing rate of each segment is used as the input feature vector for the pattern classifier. The support vector machine (SVM) classification and recognition algorithm is adopted to classify and train numerous intrusion data and save the model parameters. In unknown intrusion events, the feature vector is extracted and fed into the trained SVM model to realize high-efficiency and high-accuracy pattern recognition. A Michelson interferometer-based fiber perimeter security system is developed and a 2-km-long fiber optic cable is installed in the outdoor fence for experimental verification; 120 groups are used with a total of 600 experiments being performed under five different cases: shearing cable, climbing fence, swaying fence, tapping cable, and no intrusion. Experimental results confirm that the proposed method can quickly and accurately identify the tested types of common event signals; the average recognition rate reaches 97% and the response time is up to 0.1 s.