Abstract
Temperature sensing has seen significant advancements across modern scientific applications. While fiber-optic temperature sensors (FOTS) offer multiple advantages, achieving a synergistic combination of cost-effectiveness, compact dimensions, and high integration remains a critical challenge for practical implementations. This work presents an artificial intelligence (AI)-driven solution featuring a miniaturized, cost-efficient fiber-optic leaky-mode temperature speckle sensor with integrated optoelectronic transceiver functionality. The system employs a fiber microsphere at the emission end for light source coupling and collection, while a fiber taper at the detection end facilitates modal leakage from the fiber core. Temperature-sensitive material integrated within the sensing region modulates the refractive index (RI) in response to environmental temperature variations, thereby altering the specklegram characteristics of the leaked modes. A convolutional vision transformer (ConViT)-based regression model was trained to establish quantitative mapping between speckle image features and temperature measurements. Comparative analysis demonstrated superior performance against convolutional neural network (CNN), vision transformer (Vit), and modified CNN architectures, achieving a median absolute error (MedAE) of 0.256 °C. This integrated approach effectively addresses the critical triad of cost reduction, miniaturization, and functional integration, potentially advancing the development of fiber-optic thermometry.
Funding Information
  • National Key Research and Development Program of China (2024YFC3015200)
  • National Natural Science Foundation of China (62105020)
  • Shenzhen Science and Technology Program (KJZD20230923114310021)
  • Fundamental Research Funds for the Central Universities (00007753)

This publication has 21 references indexed in Scilit: