Scholarlink

<p>Classification of edible oil is the basis for ensuring food authenticity and quality control. This study explores the feasibility of classifying edible oils through terahertz time-domain spectroscopy (THz-TDS) in conjunction with pattern recognition methods. Two classification tasks were established: one is fatty acid type discrimination and the other is oil type discrimination. Five representative classification models, partial least squares discriminant analysis (PLS-DA), K-Nearest Neighbors (KNN), decision tree (DT), random forest (RF), and Adaptive Boosting (Adaboost), were implemented and evaluated using full-spectrum and feature-optimized datasets. The experimental results show that KNN, RF, and AdaBoost all exhibited perfect classification, with 100% accuracy on the two sets (training and prediction). The results show that the combination of THz spectroscopy and pattern recognition modeling provides a reliable and offers a robust and rapid approach and nondestructive classification of edible oils. The method can be applied to food quality evaluation and adulteration detection.</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

Rapid and Non-contact Classification of Edible Oils Using Terahertz Time-Domain Spectroscopy Combined with Pattern Recognition Techniques

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Xiaoyan Geng,
  • Leijun Xu,
  • Jihong Deng,
  • Dengmin Li,
  • Shengqi Zhang,
  • Hui Xiao,
  • Hui Jiang

摘要

Classification of edible oil is the basis for ensuring food authenticity and quality control. This study explores the feasibility of classifying edible oils through terahertz time-domain spectroscopy (THz-TDS) in conjunction with pattern recognition methods. Two classification tasks were established: one is fatty acid type discrimination and the other is oil type discrimination. Five representative classification models, partial least squares discriminant analysis (PLS-DA), K-Nearest Neighbors (KNN), decision tree (DT), random forest (RF), and Adaptive Boosting (Adaboost), were implemented and evaluated using full-spectrum and feature-optimized datasets. The experimental results show that KNN, RF, and AdaBoost all exhibited perfect classification, with 100% accuracy on the two sets (training and prediction). The results show that the combination of THz spectroscopy and pattern recognition modeling provides a reliable and offers a robust and rapid approach and nondestructive classification of edible oils. The method can be applied to food quality evaluation and adulteration detection.