In order to accurately identify the types of intrusion signals in the distributed optical fiber early warning system, this paper proposes an optical fiber intrusion signal feature extraction and recognition algorithm based on Fourier decomposition method (FDM) and permutation entropy noise reduction method. First, the FDM is used to decompose the fiber intrusion signal into a number of intrinsic frequency band functions (FIBF). Then, the permutation entropy of each FIBF component is calculated, the sensitivity of permutation entropy to noise is used to screen qualified FIBF, and the signal is reconstructed. Finally, the approximate entropy and energy of the reconstructed signal are calculated and a two-dimensional feature vector is constructed, which is sent to the support vector machine for training and recognizes the fiber intrusion signal. Experimental results show that the algorithm can effectively identify three types of optical fiber intrusion signals: tapping, trot and passing, with an average recognition accuracy of 93.33%.