Scholarlink

<p>The quality of animal protein-based foods, such as meat, seafood, milk, and eggs, is critical to human nutrition, health and safety. Yet, as global demand rises, so do concerns about fraudulent practices and quality deterioration during storage. This review evaluates the effectiveness of nondestructive techniques for protein composition analysis, which can be used to assess the quality of animal protein-based foods, specifically the freshness and authenticity across various food types. These methods include near-infrared spectroscopy (NIR), Fourier-transform near-infrared spectroscopy (FT-NIR), mid-infrared spectroscopy (MIR), ultraviolet-visible spectroscopy (UV-Vis), Raman spectroscopy, laser-induced breakdown spectroscopy (LIBS), hyperspectral imaging (HSI), as well as digital and thermal imaging techniques. Unlike conventional methods, these nondestructive techniques provide rapid results and require minimal sample preparation. Many of these techniques, including NIR, FT-NIR, MIR, and UV-Vis, are versatile and can be used to analyze liquid and solid samples, such as milk, eggs, and meat. Other techniques, such as HSI, digital, and thermal-imaging systems, are more suitable for solid samples such as seafoods, meat, and milk powder. Commonly used multivariate analysis, machine learning, and deep-learning models integrated with these nondestructive methods for prediction and classification applications are also discussed, highlighting their superior speed, efficiency, and accuracy compared to conventional techniques. Using data from these analyses, machine learning models can identify patterns in protein composition and detect anomalies that may indicate quality deviations, loss of freshness, or adulteration. This review also presents detailed discussions on data processing, feature extraction, model training, model optimization techniques, and prediction approaches. Additionally, methods for improving data processing and cross-validation steps are identified to enhance the effectiveness and generalizability of machine and deep learning models. This is imperative for making them more robust and for the successful application of these tools to wider varieties of datasets, including unseen test data.</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

Machine learning-enabled nondestructive quality analysis of animal protein-based foods: a comprehensive review

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月9日

查看全文
  • Olusola Olagunju,
  • Michael Stump,
  • Yonghui Li

摘要

The quality of animal protein-based foods, such as meat, seafood, milk, and eggs, is critical to human nutrition, health and safety. Yet, as global demand rises, so do concerns about fraudulent practices and quality deterioration during storage. This review evaluates the effectiveness of nondestructive techniques for protein composition analysis, which can be used to assess the quality of animal protein-based foods, specifically the freshness and authenticity across various food types. These methods include near-infrared spectroscopy (NIR), Fourier-transform near-infrared spectroscopy (FT-NIR), mid-infrared spectroscopy (MIR), ultraviolet-visible spectroscopy (UV-Vis), Raman spectroscopy, laser-induced breakdown spectroscopy (LIBS), hyperspectral imaging (HSI), as well as digital and thermal imaging techniques. Unlike conventional methods, these nondestructive techniques provide rapid results and require minimal sample preparation. Many of these techniques, including NIR, FT-NIR, MIR, and UV-Vis, are versatile and can be used to analyze liquid and solid samples, such as milk, eggs, and meat. Other techniques, such as HSI, digital, and thermal-imaging systems, are more suitable for solid samples such as seafoods, meat, and milk powder. Commonly used multivariate analysis, machine learning, and deep-learning models integrated with these nondestructive methods for prediction and classification applications are also discussed, highlighting their superior speed, efficiency, and accuracy compared to conventional techniques. Using data from these analyses, machine learning models can identify patterns in protein composition and detect anomalies that may indicate quality deviations, loss of freshness, or adulteration. This review also presents detailed discussions on data processing, feature extraction, model training, model optimization techniques, and prediction approaches. Additionally, methods for improving data processing and cross-validation steps are identified to enhance the effectiveness and generalizability of machine and deep learning models. This is imperative for making them more robust and for the successful application of these tools to wider varieties of datasets, including unseen test data.