Key Indicator Detection and Authenticity Identification of Beer Based on Near-Infrared Spectroscopy Combined with...

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Machine Learning

Background:

• Traditional beer detection methods face limitations in speed and accuracy.
• Verifying beer authenticity and quality requires robust analytical techniques.

Purpose of the Study:

• To develop a rapid Near-Infrared Spectroscopy (NIRS)-based method for detecting key beer indicators and verifying authenticity.
• To explore and optimize Single-task (STL) and Multi-task learning (MTL) strategies for enhanced spectral analysis.

Main Methods:

• Variable Importance in Projection (VIP) for wavelength selection.
• Deep spectral feature extraction using Convolutional Neural Network (CNN)-Multi-Head Attention (MHA), Long Short-Term Memory (LSTM)-MHA, and hybrid CNN-LSTM-MHA networks.
• Bayesian Optimization Algorithm for hyperparameter tuning in both STL and MTL frameworks.
• Development of Partial Least Squares Regression (PLSR), Support Vector Machine Regression (SVMR), and Partial Least Squares Discriminant Analysis (PLSDA) models.

Main Results:

• The Multi-task learning (MTL)-based CNN-LSTM-MHA network demonstrated superior generalization by effectively learning shared features.
• High accuracy was achieved for alcohol content (R2=0.996, rRMSE=2.024%) and original wort concentration (R2=0.997, rRMSE=2.515%) in the validation set.
• Independent testing confirmed excellent performance (R2=0.995 for alcohol, R2=0.991 for wort concentration) with 100% classification accuracy across all datasets.

Conclusions:

• The proposed NIRS method with deep learning offers an efficient and accurate solution for real-time beer quality detection.
• This technique supports effective beer market regulation and enhances quality control in production processes.
• The study highlights the potential of MTL strategies in spectral analysis for complex food matrices.