AirNet: A Deep Learning-Driven Auto Baseline Correction Algorithm Balancing Global Smoothness and Local Fidelity

Area of Science:

• Spectroscopy
• Chemometrics
• Machine Learning

Background:

• Baseline correction is crucial for accurate Raman/SERS spectral analysis.
• Existing machine learning methods lack generalizability due to poorly understood parameters.

Purpose of the Study:

• To develop an automated baseline correction algorithm, AirNet, that balances global smoothness and local fidelity.
• To improve the accuracy and generalizability of Raman spectral preprocessing.

Main Methods:

• AirNet integrates deep learning (ResUnet) with chemometrics.
• It employs a dynamic and robust adaptive iteratively reweighted penalized least squares (Dr-airPLS) method.
• An adaptive multi-indicator strategy optimizes smoothing parameters.

Main Results:

• AirNet outperforms existing algorithms (airPLS, OP-airPLS, DIRAS) in accuracy and generalizability.
• Achieves a processing speed of 0.3 seconds per spectrum.
• Demonstrates consistent correction across homologous spectra with a segmented fitting strategy.

Conclusions:

• AirNet provides high-fidelity Raman spectral information for reliable spectrum-structure correlation.
• The algorithm offers improved accuracy, generalizability, and speed in baseline correction.
• Its robust design addresses challenges with varying background gradients.