Speckle-decoded temperature-insensitive strain identification in a multimode optical fiber

Fiber-optic sensing systems are significant tools for measuring various physical or biochemical parameters. However, temperature cross-sensitivity prevents accurate recognition of the target input signal when optical sensors are applied in practical scenarios. Herein, leveraging a deep learning algorithm, a speckle-decoded temperature-insensitive strain sensor is proposed and experimentally demonstrated. Scattering patterns are utilized to estimate the axial strain since the external force could change the coherent superposition of the amplitudes of propagating modes. The experimental results show that the recognition accuracy of the sensing system based on a classification model can reach 99.28% within a wide strain range of 0-0.3 N in the presence of temperature cross talk. In addition, the strain prediction demonstrates an average root-mean-square error of 1.02 N%. Such an intelligent speckle sensing strategy has the potential to broaden the applications of fiber-optic sensors in various engineering applications.