High performance few-mode fiber-based light field direction sensing system using deep convolutional neural network:...

Area of Science:

• Optics and Photonics
• Machine Learning for Sensing
• Fiber Optic Sensors

Background:

• Precise light field direction sensing is crucial for 3D imaging, remote sensing, and target tracking.
• Optical fibers offer high sensitivity, compact size, and electromagnetic interference resistance for sensing applications.

Purpose of the Study:

• To investigate the variation characteristics of few-mode fiber output speckle patterns in response to incident light field direction changes.
• To develop a novel system for high-performance light field direction sensing using few-mode fibers and deep learning.

Main Methods:

• Simulated and analyzed few-mode fiber transmission characteristics to correlate output speckle with incident light direction (±6°).
• Proposed and constructed a deep convolutional neural network (CNN), the Fiber Speckle Demodulation Network (FSDNET), to map light field direction to speckle patterns.
• Developed and tested a light field direction sensing system integrating the few-mode fiber and FSDNET.

Main Results:

• Theoretical simulations showed a mean absolute error (MAE) of 0.01° for light field direction sensing.
• The FSDNET achieved an MAE of 0.0389° for known directions and 0.0570° for unknown directions in the experimental setup.

Conclusions:

• The proposed few-mode fiber sensing system combined with the FSDNET demonstrates high performance in accurately sensing light field direction.
• This approach offers a promising new method for precise light field direction detection in various applications.