Rapid Detection of Adulteration in Birch Sap Using Near-Infrared Spectroscopy Coupled With Chemometrics

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Birch sap adulteration poses a quality control challenge for the beverage industry.
• Traditional methods for detecting adulteration are often time-consuming and destructive.
• Developing rapid, non-destructive analytical techniques is crucial for ensuring product authenticity.

Purpose of the Study:

• To develop a green analytical approach using near-infrared (NIR) spectroscopy for rapid screening of birch sap adulteration.
• To evaluate the effectiveness of various spectral preprocessing techniques on model performance.
• To establish a robust and efficient method for both qualitative and quantitative analysis of birch sap purity.

Main Methods:

• Acquisition of NIR spectral data from birch sap samples.
• Application of spectral preprocessing techniques: Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), first derivative, and Savitzky-Golay (SG) smoothing.
• Development of a dung beetle optimization algorithm-enhanced support vector machine (DBO-SVM) for qualitative analysis (discrimination).
• Implementation of a hybrid deep learning model (CNN-transformer-ECA) for quantitative analysis (adulteration concentration prediction).
• Optimization of the deep learning model using network slimming (NS) for efficiency.

Main Results:

• The DBO-SVM model achieved 98% discrimination accuracy for pure versus adulterated birch sap after SNV preprocessing.
• The CNN-transformer-ECA hybrid model demonstrated high accuracy in predicting adulteration concentration, with R²p values of 0.9482 (water dilution) and 0.9505 (sucrose solution).
• The optimized lightweight model showed excellent fitting and generalization performance, suitable for deployment on embedded platforms.

Conclusions:

• NIR spectroscopy, combined with advanced machine learning and deep learning models, provides an effective and rapid method for birch sap adulteration screening.
• The developed DBO-SVM and CNN-transformer-ECA models offer a faster, cheaper, and non-destructive alternative to traditional quality control methods.
• This study lays the groundwork for applying NIR spectroscopy to monitor the quality of related liquid beverages, ensuring consumer trust and product integrity.