In order to reduce the influence of noise in the fiber optic perimeter security signal on the classification results and improve the accuracy and operating efficiency of signal classification, in this paper a classification method combine Correlation Variational Mode Decomposition (CVMD), Dung Beetle Optimizer (DBO), and Support Vector Machine (SVM) was proposed. CVMD was used to remove the noise component in the original signal. The energy, energy entropy and kurtosis of the denoised signal were extracted as feature vectors. The DBO algorithm was adopted to optimize the SVM to obtain the best penalty factor and kernel function parameters. The DBO-SVM classification model was constructed. A perimeter security system based on phase-sensitive optical time-domain reflectometry (Φ-OTDR) technology was built to collect four types of signals: climbing, knocking, stepping and non-intrusion. The experimental results show that the classification accuracy of CVMD-DBO-SVM is higher than that of CVMD-PSO-SVM and CVMD-GA-SVM, reaching 98.75%. At the same time, the running time is shorter and the overall performance is the best.