Raman spectral classification algorithm of cephalosporin based on VGGNeXt

Area of Science:

• Spectroscopy
• Chemometrics
• Machine Learning

Background:

• Deep learning models for Raman spectral classification often lack generalizability across different instruments.
• Existing models trained on data from a single device may fail when applied to spectra from other spectrometers.

Purpose of the Study:

• To develop a robust deep learning algorithm for classifying cephalosporin Raman spectra.
• To address the challenge of model transferability and instrument standardization in Raman spectroscopy.

Main Methods:

• Established a database of six cephalosporin Raman spectra.
• Proposed and implemented the VGGNeXt classification algorithm, inspired by ConvNeXt and Swin-T, improving upon VGG.
• Trained the model on high-resolution spectra from a benchtop spectrometer and tested on low-resolution spectra from a portable spectrometer.

Main Results:

• The VGGNeXt algorithm outperformed comparative algorithms in all tested scenarios.
• Achieved 100% accuracy on full and halved datasets after preprocessing.
• Maintained 99.9% accuracy with only 10 data points per class, demonstrating high robustness.

Conclusions:

• The VGGNeXt model demonstrates significant potential for reliable cephalosporin Raman spectral classification across diverse instruments.
• The proposed methods contribute to solving issues of model transferability and instrument standardization in Raman spectroscopy.
• The algorithm exhibits strong robustness and high accuracy, even with limited or varied spectral data.