Optical fiber vibration signal recognition scheme based on CTCM-SVMD and KOA-CNN algorithms

To enhance the detection accuracy of perimeter intrusion events in phase-sensitive optical time-domain reflectometry–based distributed fiber-optic acoustic sensing systems, we propose a detection scheme. The scheme combines a successive variational mode decomposition (SVMD) denoising algorithm optimized by the competition of tribes and cooperation of members (CTCM) algorithm with a convolutional neural network (CNN) optimized by the Kepler optimization algorithm (KOA). The CTCM-SVMD algorithm adaptively determines the optimal quadratic penalty factor α by minimizing the envelope entropy, enabling the extraction of intrinsic mode functions (IMFs) that better match the signal characteristics. Effective IMFs are selected using Pearson’s correlation coefficient for signal reconstruction. Compared with conventional methods such as SVMD, ensemble empirical mode decomposition, and variational mode decomposition, compared with the conventional SVMD algorithm in which the signal-to-noise ratio is 13.34 dB and the root mean square error is 0.0927. The proposed method improves the signal-to-noise ratio to 17.39 dB and reduces the root mean square error to 0.0551 and at the same time effectively mitigates the pattern mixing. Subsequently, a mixed feature dataset is extracted from the denoised signals, and the CNN hyperparameters, including the learning rate, the size of the convolutional kernel, and the number of neurons, are optimized using KOA for precise classification. By processing the vibration signals obtained from the water pipeline monitoring system, it is shown that the scheme proposed in this article has a detection rate of 99%, and the average recognition rate is improved by 6% compared with traditional CNN methods. The recognition rate is 97.1% in identifying leaks, a common intrusion in leaking. This is superior to existing methods in recognizing intrusion vibration signals in real environments.