Scholarlink

<p>Polysaccharides, saponins, and flavonoids are the major bioactive compounds of <i>Polygonatum sibiricum</i> and serve as key indicators for evaluating the quality of nine-steamed and nine-dried products. The nine-steaming process exerts a significant influence on both the contents of these bioactive compounds and the pharmacological efficacy of <i>P. sibiricum</i>. Therefore, developing a rapid and nondestructive method for component analysis and process end-point determination is essential for quality control and standardized production of medicinal and edible homology materials. In this study, hyperspectral imaging technology was employed to acquire spectral information of <i>P. sibiricum</i> at different steaming stages. A dual-branch CNN encoder framework, termed Dual-TCNet, was proposed for the modeling of spectral data. Hyperspectral data were integrated with the deep neural network architecture to enable nondestructive analysis of the major bioactive compounds of <i>P. sibiricum</i> (polysaccharides, saponins, and flavonoids) and accurate determination of the process end-point in the nine-time steaming and drying procedure. The model employed a convolutional neural network (CNN) to extract local spectral features and an encoder module to capture global spectral features. To achieve feature integration across multiple scales, a multi-head attention mechanism was applied. To clearly demonstrate the comparative performance of the proposed approach, the experimental results were systematically compared with widely used benchmark models, including random forest (RF), partial least squares (PLS), and one-dimensional convolutional neural network (1D-CNN). The Dual-TCNet model achieved a relatively high Rp of approximately 0.92 on the test set, outperforming the other models. Furthermore, the Dual-TCNet model achieved an accuracy exceeding 98% in process end-point determination during the nine-steamed and nine-dried processing, outperforming the other models. The research results indicated that the Dual-TCNet model could achieve rapid nondestructive detection of the active components in nine-steamed and nine-dried <i>P. sibiricum</i>. In addition, it could accurately determine the steaming end-point, providing a new technical approach for quality control and standardized evaluation of traditional medicinal and edible processing.</p> Graphical Abstract <p></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

A dual-branch encoder model for nondestructive detection and end-point discrimination of active components in nine-steamed and nine-dried Polygonatum sibiricum using hyperspectral technology

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Anguo Xie,
  • Bingqian Liu,
  • Han Wu,
  • Yunhong Liu

摘要

Polysaccharides, saponins, and flavonoids are the major bioactive compounds of Polygonatum sibiricum and serve as key indicators for evaluating the quality of nine-steamed and nine-dried products. The nine-steaming process exerts a significant influence on both the contents of these bioactive compounds and the pharmacological efficacy of P. sibiricum. Therefore, developing a rapid and nondestructive method for component analysis and process end-point determination is essential for quality control and standardized production of medicinal and edible homology materials. In this study, hyperspectral imaging technology was employed to acquire spectral information of P. sibiricum at different steaming stages. A dual-branch CNN encoder framework, termed Dual-TCNet, was proposed for the modeling of spectral data. Hyperspectral data were integrated with the deep neural network architecture to enable nondestructive analysis of the major bioactive compounds of P. sibiricum (polysaccharides, saponins, and flavonoids) and accurate determination of the process end-point in the nine-time steaming and drying procedure. The model employed a convolutional neural network (CNN) to extract local spectral features and an encoder module to capture global spectral features. To achieve feature integration across multiple scales, a multi-head attention mechanism was applied. To clearly demonstrate the comparative performance of the proposed approach, the experimental results were systematically compared with widely used benchmark models, including random forest (RF), partial least squares (PLS), and one-dimensional convolutional neural network (1D-CNN). The Dual-TCNet model achieved a relatively high Rp of approximately 0.92 on the test set, outperforming the other models. Furthermore, the Dual-TCNet model achieved an accuracy exceeding 98% in process end-point determination during the nine-steamed and nine-dried processing, outperforming the other models. The research results indicated that the Dual-TCNet model could achieve rapid nondestructive detection of the active components in nine-steamed and nine-dried P. sibiricum. In addition, it could accurately determine the steaming end-point, providing a new technical approach for quality control and standardized evaluation of traditional medicinal and edible processing.

Graphical Abstract