Hyperspectral imaging-ensemble learning algorithm-based rapid quantitative detection of pH and amino acid nitrogen in...

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Machine Learning

Background:

• Fermented grains (Zaopei) are crucial in Baijiu production, with pH and amino acid nitrogen (AAN) as key quality indicators.
• Traditional methods for monitoring these parameters are often time-consuming and labor-intensive.

Purpose of the Study:

• To develop a rapid and accurate method for quantitative detection of pH and AAN during fermentation using hyperspectral imaging (HSI).
• To evaluate the effectiveness of various spectral preprocessing and feature selection techniques combined with ensemble learning algorithms for predicting these parameters.

Main Methods:

• Hyperspectral imaging (HSI) was employed to collect spectral data from fermented grains.
• Standard Normal Variate (SNV) was used for spectral preprocessing.
• Feature selection methods including Competitive Adaptive Reweighted Sampling (CARS), Principal Component Analysis (PCA), Interval Variable Selection Optimization (IVSO), and PCA-IVSO were evaluated.
• Ensemble learning models such as Decision Tree (DT), XGBoost, and DT-Adaptive Boosting (AdaBoost) were constructed for prediction.

Main Results:

• The PCA-IVSO strategy effectively reduced data dimensionality and improved prediction accuracy.
• The DT-AdaBoost model with PCA-IVSO achieved excellent performance for pH prediction (R_P² = 0.9998, RMSEP = 0.0029).
• The DT-AdaBoost model utilizing IVSO-selected features demonstrated superior performance for AAN prediction (R_P² = 0.9999, RMSEP = 0.0003).

Conclusions:

• HSI coupled with ensemble learning provides a robust and efficient approach for real-time quantitative detection and visual mapping of pH and AAN in fermented grains.
• This integrated technique holds substantial potential for enhancing quality control and process optimization in the Baijiu brewing industry.