In order to improve the recognition accuracy of intrusion vibration events by distributed optical fiber acoustic sensing system based on phase-sensitive optical time domain reflectometer, a recognition method of fiber optic vibration signals using endpoint detection and signal recombination is proposed. The method first uses EMD_PCC-based denoising to denoise the signal, then performs endpoint detection on the denoised signal, reorganizes the detected vibration signal and extracts the frequency domain information of the vibration signal, uses a dual-input multi-scale convolutional neural network to extract the time domain features and frequency domain features of the vibration signal respectively, and finally uses a support vector machine for classification. The experiment proves that the recognition method can quickly complete the training of the recognition model and effectively recognize the intrusion vibration signals collected in the actual environment, and the recognition accuracy of the intrusion signals can reach 94.8%.