Scholarlink

<p>Measuring beer contents, especially initial gravity, is crucial for quality assurance in brewing. This work describes an integrated system to measure beer's original gravity that uses spectroscopy, spectral preprocessing methods, wavelength selection, and sophisticated machine learning (ML) models. Original gravity, which represents the sugar level of the wort prior to fermentation, is directly proportional to the beer's eventual alcohol content. The captured spectral data were preprocessed before being fed into various ML&#xa0;algorithms to estimate beer content. The minimal-redundancy-maximal-relevance (mRMR) technique was used to find the optimal wavelengths, thereby significantly enhancing the precision of the models.</p><p>Regression results showed a high coefficient of determination (R<sup>2</sup> = 0.998), low root mean square error (RMSE = 0.123), and an excellent ratio of performance to deviation (RPD = 21.75). Leave-One-Out Cross-Validation (LOOCV) confirmed robust performance (R<sup>2</sup> = 0.998, RMSE = 0.112). Classification models achieved an average Matthews Correlation Coefficient (MCC) of 95.6% and a Kappa score of 95.2%. These findings indicate that the non-destructive technology, combined with mRMR, provides a dependable and efficient way of determining beer levels.</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

Non-destructive beer gravity measurement using spectroscopy and machine learning with mRMR-based wavelength selection

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Kamini G Panchbhai,
  • Madhusudan G Lanjewar,
  • Panem Charanarur

摘要

Measuring beer contents, especially initial gravity, is crucial for quality assurance in brewing. This work describes an integrated system to measure beer's original gravity that uses spectroscopy, spectral preprocessing methods, wavelength selection, and sophisticated machine learning (ML) models. Original gravity, which represents the sugar level of the wort prior to fermentation, is directly proportional to the beer's eventual alcohol content. The captured spectral data were preprocessed before being fed into various ML algorithms to estimate beer content. The minimal-redundancy-maximal-relevance (mRMR) technique was used to find the optimal wavelengths, thereby significantly enhancing the precision of the models.

Regression results showed a high coefficient of determination (R2 = 0.998), low root mean square error (RMSE = 0.123), and an excellent ratio of performance to deviation (RPD = 21.75). Leave-One-Out Cross-Validation (LOOCV) confirmed robust performance (R2 = 0.998, RMSE = 0.112). Classification models achieved an average Matthews Correlation Coefficient (MCC) of 95.6% and a Kappa score of 95.2%. These findings indicate that the non-destructive technology, combined with mRMR, provides a dependable and efficient way of determining beer levels.