High-quality full stokes polarimetric spectroscopy reconstruction using a model-compensated physics-informed neural network for channeled spectropolarimetry

Area of Science:

• Optical Physics
• Data Science
• Spectroscopy

Background:

• Channeled spectropolarimetry (CSP) offers simultaneous full Stokes spectra acquisition.
• Existing CSP methods face challenges with systematic errors and simplified physical models, limiting reconstruction accuracy.
• Reliability issues hinder the application of CSP in dynamic and complex optical measurements.

Purpose of the Study:

• To introduce a novel model-compensated physics-informed neural network (MC-PINN) for enhanced CSP spectral reconstruction.
• To address limitations of traditional CSP models and improve data reliability.
• To achieve high-quality spectral reconstruction using deep learning and improved physical models.

Main Methods:

• Developed a model-compensated physics-informed neural network (MC-PINN) integrating deep learning with an improved physical model.
• Implemented a prediction network to approximate data and physical laws.
• Incorporated a compensation network to correct model inaccuracies and discrepancies.

Main Results:

• MC-PINN significantly reduces errors stemming from physical model inaccuracies.
• Achieved high-quality spectral reconstruction even with limited and noisy data.
• Demonstrated superior accuracy and robustness compared to traditional methods in simulations and experiments.

Conclusions:

• MC-PINN enhances the accuracy and robustness of CSP spectral reconstruction.
• The method effectively reconstructs high-frequency spectral features and complex polarization states.
• MC-PINN shows significant potential for improving data reliability and expanding CSP applications in challenging scenarios.