State Key Laboratory of Information Photonics and Optical Communications, School of Electronic Engineering, Beijing University of Posts and Telecommunications (BUPT), Beijing 100876, China.
Beijing Key Laboratory of Space-Round Interconnection and Convergence, Beijing 100876, China.
Sensors (Basel). 2023 Mar 31;23(7):3637. doi: 10.3390/s23073637.
In this paper, aiming at a large infrastructure structural health monitoring network, a quaternion wavelet transform (QWT) image denoising algorithm is proposed to process original data, and a depth feedforward neural network (FNN) is introduced to extract physical information from the denoised data. A Brillouin optical time domain analysis (BOTDA)-distributed sensor system is established, and a QWT denoising algorithm and a temperature extraction scheme using FNN are demonstrated. The results indicate that when the frequency interval is less than 4 MHz, the temperature error is kept within ±0.11 °C, but is ±0.15 °C at 6 MHz. It takes less than 17 s to extract the temperature distribution from the FNN. Moreover, input vectors for the Brillouin gain spectrum with a frequency interval of no more than 6 MHZ are unified into 200 input elements by linear interpolation. We hope that with the progress in technology and algorithm optimization, the FNN information extraction and QWT denoising technology will play an important role in distributed optical fiber sensor networks for real-time monitoring of large-scale infrastructure.
本文针对大型基础设施结构健康监测网络，提出了一种四元数小波变换（QWT）图像去噪算法来处理原始数据，并引入深度前馈神经网络（FNN）从去噪数据中提取物理信息。建立了布里渊光时域分析（BOTDA）分布式传感器系统，对 QWT 去噪算法和基于 FNN 的温度提取方案进行了验证。结果表明，当频率间隔小于 4MHz 时，温度误差保持在±0.11°C 以内，但在 6MHz 时为±0.15°C。从 FNN 中提取温度分布所需的时间不到 17s。此外，通过线性插值，将频率间隔不超过 6MHz 的布里渊增益谱的输入向量统一为 200 个输入元素。我们希望随着技术和算法优化的进步，FNN 信息提取和 QWT 去噪技术将在用于大型基础设施实时监测的分布式光纤传感器网络中发挥重要作用。
