Fiber-Optic Sensor Spectrum Noise Reduction Based on a Generative Adversarial Network

Area of Science:

• Optoelectronics and Photonics
• Signal Processing
• Artificial Intelligence in Sensing

Background:

• Reducing noise in fiber-optic sensor spectra is crucial for accurate measurements.
• Existing denoising methods like wavelet transform (WT) and empirical mode decomposition (EMD) have limitations.

Purpose of the Study:

• To develop and evaluate a deep-learning-based denoising method for fiber-optic sensor spectra.
• To enhance the signal-to-noise ratio (SNR) and accuracy of fiber-optic sensing data.

Main Methods:

• Pre-processing sensor spectra into 2D images.
• Training a cycle-consistent generative adversarial network (Cycle-GAN) model.
• Evaluating performance on simulated spectra from FPI, FBG, chirped FBG, and FBG pair sensors.

Main Results:

• Achieved SNR improvement up to 13.71 dB compared to traditional methods.
• Reduced RMSE by up to three times and maintained R2 ≥ 99.70% with original signals.
• Demonstrated excellent linearity (R2 of 99.95%) for multimode noise reduction in temperature response.

Conclusions:

• The proposed Cycle-GAN denoising method effectively reduces various noise types in fiber-optic sensing.
• This approach enhances the practicality and reliability of fiber-optic sensors for specialized applications.
• The method shows significant advantages over traditional denoising techniques.