Deep-Learning-Based Broadband Lightsource X-ray Absorption Spectroscopy Using Photon-Counting Detector.

Area of Science:

• Physics
• Materials Science
• Computer Science

Background:

• Photon-counting detectors offer high-resolution, multienergy X-ray spectroscopy.
• High photon flux causes spectral distortions (pile-up, polarization) in X-ray spectroscopy.
• Accurate spectral data is crucial for material analysis.

Purpose of the Study:

• To develop a deep learning framework for correcting spectral distortions in broadband lightsource X-ray absorption spectroscopy (BL-XAS).
• To enhance the spectral integrity of X-ray absorption spectroscopy data acquired under high photon flux conditions.

Main Methods:

• Proposed Spectral-Transformer, a deep learning model fusing spectral and tube current data.
• Utilized a bimodal mapping mechanism and physics-informed loss functions for distortion correction.
• Developed a cascade simulation framework for generating high-fidelity training data, incorporating pile-up and polarization effects.

Main Results:

• The Spectral-Transformer framework successfully corrected spectral distortions.
• Corrected spectra demonstrated exceptional fidelity, reducing KL divergence to 1.0 × 10^-4.
• Material classification accuracy improved significantly from 86.2% to 95.5% using corrected spectra.

Conclusions:

• The Spectral-Transformer is an effective deep learning solution for spectral distortion correction in BL-XAS.
• Enhanced spectral integrity leads to improved analytical performance and material classification.
• This approach advances the application of photon-counting detectors in high-flux X-ray spectroscopy.