Intelligent Optical Fiber-Integrated Near-Infrared Polarimeter Based on Upconversion Nanoparticles

The near-infrared (NIR) light polarization metrology is crucial in a broad area of applications, such as optical sensing, imaging, and communications. The conventional measurement of NIR light polarization requires bulky polarization optics by the spatial or temporal splitting of the input light beam with expensive infrared detectors. Here, an intelligent fiber-integrated NIR polarimeter based on upconversion nanoparticles (UCNPs) is demonstrated. Er-doped UCNPs are synthesized in core-shell structures for bright upconversion luminescence (UCL), and the tapered multimode fiber (MMF) is designed for random Rayleigh-distributed speckles. It is demonstrated for the first time that spectra-polarization-dependent upconverted speckle patterns captured by a cost-effective visible image sensor can be accurately decoded via a deep neural network (ResNet50). The experimental results show that ResNet50 has a recognition accuracy of over 99.5% for the full-Stokes parameters of NIR light polarization, along with 100% accuracy for the light wavelength ranging from 1535 nm to 1565 nm. The developed intelligent optical sensors provide a new design paradigm for fiber-integrated multi-dimensional NIR light field detection and can be applied in fiber-optic sensors and communications networks.