Opt Lett. 2023 Apr 15;48(8):2114-2117. doi: 10.1364/OL.487049.
In the fiber Bragg grating (FBG) sensor network, the signal resolution of the reflected spectrum is correlated with the network's sensing accuracy. The interrogator determines the signal resolution limits, and a coarser resolution results in an enormous uncertainty in sensing measurement. In addition, the multi-peak signals from the FBG sensor network are often overlapped; this increases the complexity of the resolution enhancement task, especially when the signals have a low signal-to-noise ratio (SNR). Here, we show that deep learning with U-Net architecture can enhance the signal resolution for interrogating the FBG sensor network without hardware modifications. The signal resolution is effectively enhanced by 100 times with an average root mean square error (RMSE) < 2.25 pm. The proposed model, therefore, allows the existing low-resolution interrogator in the FBG setup to function as though it contains a much higher-resolution interrogator.
在光纤布拉格光栅（FBG）传感器网络中，反射光谱的信号分辨率与网络的传感精度相关。解调器决定了信号分辨率的极限，而较低的分辨率会导致传感测量的极大不确定性。此外，来自 FBG 传感器网络的多峰信号常常重叠；这增加了分辨率增强任务的复杂性，特别是当信号的信噪比（SNR）较低时。在这里，我们展示了基于 U-Net 架构的深度学习可以在不进行硬件修改的情况下增强 FBG 传感器网络的信号分辨率。信号分辨率的有效增强倍数达到 100 倍，平均均方根误差（RMSE）<2.25 pm。因此，所提出的模型允许 FBG 设备中现有的低分辨率解调器能够像包含更高分辨率解调器一样运行。
