Scholarlink

<p>Tea is a very popular non alcoholic beverage throughout the world. The quality of tea leaf is assessed to ensure the taste, aroma and its ability to reduce the risk of some health issues. The polyphenol content of tea leaf greatly demands the enrichment of tea quality. The present work shows the implementation of a quality estimator of tea leaf samples based on a very efficient analytical technique - Fourier Transform Near Infrared (FT-NIR) spectroscopy. Folin Ciocalteu’s (FC) phenol reagent is used while deploying the standard method of ISO 14502-1 (ISO, 2005) to measure reference FC values for the spectral data set. The reflectance spectral data for tea leaf samples in powdered form are preprocessed by MSC and SNV techniques. A Convolutional Neural Network (CNN) model is framed to extract the features from these two types of preprocessed spectral data of the tea samples separately. In this model effective features are deeply processed by several convolution layers and the loss function- mean squared error (MSE) is minimized greatly by the neural network. By train test split, the predictions of the total polyphenol content (FC values) are made and compared to the true value of this parameter for both MSC and SNV processed data. The efficiency of this estimator is established by evaluating the performance metrics MSE, MAE and correlation score (R<sup>2</sup>).</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

Convolutional Neural Network Based Deep Learning Approach of Polyphenol Content Prediction for Fresh Tea Leaves Using NIR Spectroscopy

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • Reshma Sengupta,
  • Anil Kumar Bag,
  • Bipan Tudu,
  • Rajib Bandyopadhyay

摘要

Tea is a very popular non alcoholic beverage throughout the world. The quality of tea leaf is assessed to ensure the taste, aroma and its ability to reduce the risk of some health issues. The polyphenol content of tea leaf greatly demands the enrichment of tea quality. The present work shows the implementation of a quality estimator of tea leaf samples based on a very efficient analytical technique - Fourier Transform Near Infrared (FT-NIR) spectroscopy. Folin Ciocalteu’s (FC) phenol reagent is used while deploying the standard method of ISO 14502-1 (ISO, 2005) to measure reference FC values for the spectral data set. The reflectance spectral data for tea leaf samples in powdered form are preprocessed by MSC and SNV techniques. A Convolutional Neural Network (CNN) model is framed to extract the features from these two types of preprocessed spectral data of the tea samples separately. In this model effective features are deeply processed by several convolution layers and the loss function- mean squared error (MSE) is minimized greatly by the neural network. By train test split, the predictions of the total polyphenol content (FC values) are made and compared to the true value of this parameter for both MSC and SNV processed data. The efficiency of this estimator is established by evaluating the performance metrics MSE, MAE and correlation score (R2).