Abstract

In order to improve the signal-to-noise ratio (SNR) of vibration sensing in the phase-sensitive optical time-domain reflectometer (φ-OTDR) system, a fiber sensing signal processing method based on the FFDNet convolutional neural network is proposed in this paper. In the network, the concept of residual learning is introduced, which involves constructing a residual mapping and utilizing multi-layer convolutional neural networks to learn the noise distribution present in the original image. The denoised result can be obtained by subtracting the learned noise from the original image. We have built a φ-OTDR system based on coherent detection, using three PZTs as simulated vibration sources and a series of experiments at 200 Hz, with each experiment simulating a single vibration event or multiple vibration events by setting different intensities. The experimental results demonstrate that the FFDNet based fiber optic sensing signal processing method enhances the SNR to 37.84 dB, 37.11 dB, and 37.31 dB, respectively, while preserving vibration signal details more effectively than wavelet denoising and Gaussian filtering techniques. The trained FFDNet model has great potential for improving the performance of the φ-OTDR system and has some practical application value.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call