Deep Learning for Temperature Sensing With Microstructure Fiber in Noise Perturbation Environment

Owing to their unique structural design and strong light-matter interaction, microstructure optical fibers (MOFs) emerge as a promising platform for optical sensing. When MOF-based sensors are perturbed, the acquired spectrum data are scrambled that significantly limits their practical application. To overcome this challenge, a functionalized MOF enabled by deep learning-based data analytics is proposed and experimentally demonstrated to realize temperature sensing in the presence of external perturbations. Under different temperature states, the trained deep learning model can effectively extract features from the collected signal data and achieve highly accurate recognition of complex speckle patterns. Experimental results show that the speckle recognition accuracy of the deep learning model reaches 98.64% and 100% in the presence and absence of external perturbations, respectively. The approach herein can be extended to other MOF-based sensing systems with regular or irregular speckle patterns for detecting various measurands. And moreover, the MOF integrated with an optical neural network possesses several desirable features such as high recognition accuracy, anti-interference capability and simplified configuration, showing great potential toward realizing the smart monitoring applications in the aspects of healthcare, industry, and environmental engineering.