Scholarlink

<p>The relationships among trace components in Baijiu base liquor are complex and diverse. An evaluation model was developed based on near-infrared (NIR) spectral data to enable rapid and convenient prediction of its quality grade. First, the weighted SPXY (WSPXY) method was employed to comprehensively consider spectral and target variable spaces for training–testing set division. Second, an adaptive multifactor feature evolutionary algorithm (AMFEA) was introduced. By integrating feature selection, variance analysis, and Spearman correlation coefficient tasks with a knowledge transfer strategy, AMFEA screened 83 features from the preprocessed base spirit spectra, which were then used as inputs for the model. Finally, an extreme gradient boosting (XGBoost) model with Bayesian parameter optimization (Optuna) was employed for grade classification of the base liquor. The results indicate that the features extracted by the AMFEA–Optuna–XGBoost algorithm effectively represent the chemical composition of the base liquor, achieving an accuracy, precision, recall, and F1-score of 95.86%, 96.62%, 95.83%, and 96.21%, respectively. The proposed method provides a reference for rapidly detecting Baijiu base liquor grades.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Near-Infrared Spectroscopy Combined With An Adaptive Multi-Factor Feature Evolution Algorithm For Identifying Base Liquor Grades

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Guiyu Zhang,
  • Yaohong Tang,
  • Rutao He,
  • Xianglin Zeng,
  • Gao Li

摘要

The relationships among trace components in Baijiu base liquor are complex and diverse. An evaluation model was developed based on near-infrared (NIR) spectral data to enable rapid and convenient prediction of its quality grade. First, the weighted SPXY (WSPXY) method was employed to comprehensively consider spectral and target variable spaces for training–testing set division. Second, an adaptive multifactor feature evolutionary algorithm (AMFEA) was introduced. By integrating feature selection, variance analysis, and Spearman correlation coefficient tasks with a knowledge transfer strategy, AMFEA screened 83 features from the preprocessed base spirit spectra, which were then used as inputs for the model. Finally, an extreme gradient boosting (XGBoost) model with Bayesian parameter optimization (Optuna) was employed for grade classification of the base liquor. The results indicate that the features extracted by the AMFEA–Optuna–XGBoost algorithm effectively represent the chemical composition of the base liquor, achieving an accuracy, precision, recall, and F1-score of 95.86%, 96.62%, 95.83%, and 96.21%, respectively. The proposed method provides a reference for rapidly detecting Baijiu base liquor grades.