Scholarlink

<p>This study investigated the application of VIS-NIR spectroscopy coupled with unsupervised machine learning algorithms for determining salt levels in toast bread slices. Samples were prepared at three salt levels: low (7–13&#xa0;g), normal (16–25&#xa0;g), and high (28–34&#xa0;g), and stored at room temperature and in the refrigerator (4&#xa0;°C) to evaluate the effect of storage conditions during 7 days. Spectral data were captured from the crust and crumb of each slice directly and through the plastic package. The data were preprocessed using MSC, SNV, and SG filters and then analyzed using unsupervised clustering algorithms. Results indicated that salt level significantly influenced spectral differences in both crumb and crust samples. Among the preprocessing methods, MSC yielded the best performance, with PC1 explaining 88–93% of the variance in crumb samples and 85–98% in crust samples. K-means clustering achieved high clustering quality, with silhouette indices of 0.712 (crumb, unpackaged-Room temperature storage) and 0.698 (crumb, packaged-Room temperature storage), and 0.723 (crust, unpackaged-Room temperature storage) and 0.726 (crust, packaged-Room temperature storage). SOM clustering confirmed these patterns, with Davies–Bouldin indices ranging from 0.62 to 0.66 across storage conditions. The artificial bee colony clustering algorithm achieved the highest performance for packaged crust (silhouette = 0.714) and crumb (0.632), demonstrating clear separation among the low-, normal-, and high-salt groups. Overall, combining VIS-NIR spectroscopy with clustering algorithms provides a practical, rapid, and non-destructive method for determining the salt level in toast bread, and potentially supporting on-site and out-of-laboratory monitoring of salt usage. The proposed approach is intended as an exploratory framework for non-destructive salt-level discrimination paving the way for future rapid assessment methods.</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

Salt-level discrimination in toast bread using handheld vis-NIR spectroscopy combined with unsupervised machine learning algorithms

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Mohammad Vahedi Tarshizi,
  • Saeid Minaei,
  • Sajad Kiani

摘要

This study investigated the application of VIS-NIR spectroscopy coupled with unsupervised machine learning algorithms for determining salt levels in toast bread slices. Samples were prepared at three salt levels: low (7–13 g), normal (16–25 g), and high (28–34 g), and stored at room temperature and in the refrigerator (4 °C) to evaluate the effect of storage conditions during 7 days. Spectral data were captured from the crust and crumb of each slice directly and through the plastic package. The data were preprocessed using MSC, SNV, and SG filters and then analyzed using unsupervised clustering algorithms. Results indicated that salt level significantly influenced spectral differences in both crumb and crust samples. Among the preprocessing methods, MSC yielded the best performance, with PC1 explaining 88–93% of the variance in crumb samples and 85–98% in crust samples. K-means clustering achieved high clustering quality, with silhouette indices of 0.712 (crumb, unpackaged-Room temperature storage) and 0.698 (crumb, packaged-Room temperature storage), and 0.723 (crust, unpackaged-Room temperature storage) and 0.726 (crust, packaged-Room temperature storage). SOM clustering confirmed these patterns, with Davies–Bouldin indices ranging from 0.62 to 0.66 across storage conditions. The artificial bee colony clustering algorithm achieved the highest performance for packaged crust (silhouette = 0.714) and crumb (0.632), demonstrating clear separation among the low-, normal-, and high-salt groups. Overall, combining VIS-NIR spectroscopy with clustering algorithms provides a practical, rapid, and non-destructive method for determining the salt level in toast bread, and potentially supporting on-site and out-of-laboratory monitoring of salt usage. The proposed approach is intended as an exploratory framework for non-destructive salt-level discrimination paving the way for future rapid assessment methods.