Online detection of Q-marker concentrations in the Xuefu Zhuyu oral liquid extraction process using a multi-source cross-scale NIR attention fusion neural network

Daolong Liu ¹, Lele Gao ², Zihao Cui ¹

⁰School of Mechanical Engineering, Shandong University, Jinan 250061, China; Key Laboratory of High Efficiency and Clean Mechanical Manufacture, Ministry of Education, Jinan 250061, China.

Spectrochimica Acta. Part a, Molecular and Biomolecular Spectroscopy +

|

May 1, 2025

View abstract on PubMed

Summary

A new Multi-Source Cross-Scale Attention Fusion Network (MSCAF-Net) improves near-infrared (NIR) spectroscopy for traditional Chinese medicine (TCM) quality control. This advanced model enhances real-time monitoring accuracy for key compounds in TCM extracts.

Area Of Science

Analytical Chemistry
Spectroscopy
Process Analytical Technology (PAT)

Background

Near-infrared (NIR) spectroscopy is crucial for real-time quality monitoring in traditional Chinese medicine (TCM) production.
Industrial environments introduce interference factors like noise and temperature fluctuations, degrading NIR model accuracy and consistency.
Existing methods struggle with spectral data instability and baseline drift, limiting reliable quality control.

Purpose Of The Study

To develop an advanced deep learning model for robust and accurate real-time prediction of quality marker (Q-Marker) concentrations in TCM extracts.
To overcome limitations of single-spectrometer NIR models by fusing data from multiple sources.
To enhance the precision and reliability of process analytical technology (PAT) in TCM manufacturing.

Main Methods

Proposed a Multi-Source Cross-Scale Attention Fusion Network (MSCAF-Net) integrating spectral data from two distinct NIR spectrometers.
Implemented a cross-scale feature extraction module and a multi-head attention mechanism for effective feature fusion.
Utilized a three-layer convolutional neural network with varying kernel sizes for regression-based Q-Marker concentration predictions.

Main Results

MSCAF-Net demonstrated superior performance in quantitative prediction of naringin, paeoniflorin, and amygdalin in Xuefu Zhuyu Oral Liquid (XZOL).
Achieved high R² values (0.9870, 0.9723, 0.8953) for the three Q-Markers, outperforming single-spectrometer and recent fusion-based methods.
The model effectively reduced signal-to-noise ratio and improved prediction accuracy and robustness.

Conclusions

MSCAF-Net offers a significant advancement for real-time quality control in TCM production by effectively fusing multi-source NIR spectral data.
The proposed network architecture enhances the robustness and accuracy of NIR spectroscopic analysis in challenging industrial environments.
This approach holds practical value for ensuring batch-to-batch consistency and quality assurance in TCM manufacturing.

Keywords:

Cross-scale feature extraction Multi-head attention mechanism NIR spectroscopy Process analytical technology Real-time monitoring and quality control Traditional Chinese Medicine extracts