Removing non-resonant background from broadband CARS using a physics-informed neural network

Area of Science:

• Spectroscopy
• Chemical Analysis
• Computational Chemistry

Background:

• Broadband Coherent Anti-Stokes Raman Scattering (BCARS) offers rapid, high-quality spectral data.
• Raw BCARS spectra contain non-resonant electronic signals that complicate chemical analysis.
• Previous methods using simulated data showed promise for Raman signal retrieval.

Purpose of the Study:

• To develop an improved computational system for BCARS data analysis.
• To accurately retrieve pure Raman signals from complex BCARS spectra.
• To enhance the reliability of chemical analyses using BCARS spectroscopy.

Main Methods:

• Developed a computational system integrating experimental laser properties and simulated susceptibilities.
• Employed a deep convolutional autoencoder network trained on system-specific data.
• Validated the method using both simulated and experimentally acquired BCARS data.

Main Results:

• The new system successfully maps susceptibility to the measured BCARS response.
• Accurate retrieval of Raman signals was demonstrated on simulated and experimental datasets.
• The computational approach effectively mitigates non-resonant background interference.

Conclusions:

• The developed computational system enhances the accuracy of Raman signal extraction from BCARS spectra.
• This method improves the quality and reliability of chemical analyses performed with BCARS.
• The system provides a robust solution for addressing spectral interferences in BCARS measurements.