Considering the complex and changeable intrusion events surrounding optical fiber perimeter and noise interference, denoising of signals collected by optical fiber perimeter systems is required. First, we propose the pelican optimization algorithm variational mode decomposition to reduce noise, and the multiscale permutation entropy decision mechanism to improve the ability to suppress modal aliasing and false components. Finally, screened signals are composed of reconstructed signals to realize the entire signal denoising. The experimental results demonstrate that signal denoising of the double Mach-Zehnder optical fiber perimeter sensing system was improved during the intrusion detection process. The denoising signal-to-noise ratio (SNR), correlation coefficient, and the mean square error of the signal after noise reduction were considered evaluation indices with regard to noise reduction performance. Compared with the existing methods of ensemble empirical mode decomposition-correlation coefficient and complementary ensemble empirical mode decomposition-correlation coefficient, the SNR of the proposed method is reduced, the correlation coefficient is clearly improved, and the root mean square error of the denoising signal is slightly reduced.