High Precision Raman Distributed Fiber Sensing Using Residual Composite Dual-Convolutional Neural Network
Haosen Guo, Jian Li, Xue Xiao-hui, Mingjiang Zhang
Abstract
Raman distributed optical fiber sensing has the unique ability to measure the spatially distributed profile of temperature that are of great interest to numerous field applications. However, the sensing performance is severely limited by the signal-to-noise ratio (SNR). The existing SNR enhancement schemes have drawbacks such as increased system complexity, degradation of sensor performance metrics such as spatial resolution, poor denoising performance, etc. Here, we report the Raman residual composite dual-convolutional neural network (RRCDNet), a novel convolutional neural network-based denoising model for one-dimensional signals specifically tailored to Raman distributed fiber sensing. The RRCDNet-enhanced Raman distributed fiber sensor system dramatically improves the temperature precision by more than a factor of 100, from 7.57°C to 0.06°C, without hardware modification or degradation of other performance metrics. At the same time, RRCDNet can also enhance other optical fiber sensor systems with one-dimensional signals, such as Rayleigh and Brillouin sensing systems.