Scholarlink

<p>Milk is a nutritional staple, yet its safety is compromised by the illegal practice of adding urea to falsely inflate protein content, with concentrations exceeding 0.5 g/L posing severe health risks. To overcome the limitations of current complex, time-consuming, and destructive detection methods, we introduce a rapid, non-destructive quantitative approach for urea detection in liquid milk using THz-TDS coupled with chemometrics. Milk-urea samples across a broad range of concentrations, from 0 to 50% (w/w), were prepared to ensure robust calibration, and the corresponding absorption coefficient spectra were collected within the 0.22–1.7 THz range. Optimization of preprocessing involved evaluating seven algorithms; the combined SNV + SG smoothing proved optimal, enhancing the PLS model’s prediction set determination coefficient (R<sup>2</sup><sub>P</sub>) to 0.9408. Subsequently, 16 quantitative regression models were developed by combining four feature selection methods (CARS, SPA, UVE, VISSA) and four regression algorithms (SVR, LSTM, RF, PSO-BP). The VISSA-RF model, which integrates VISSA dimensionality reduction with RF regression, consistently exhibited the superior predictive performance. This optimal model yielded a prediction set R<sup>2</sup><sub>P</sub> of 0.9845 and a root mean square error of prediction (RMSE<sub>P</sub>) of 0.0197, marking an accuracy increase of up to 1.38% over the 15 alternative methods. Ultimately, this study validates the high-precision and robust capability of THz spectroscopy for the non-destructive quantitative determination of urea adulterants, providing a novel and efficient technical pathway for on-line and rapid screening in dairy product safety monitoring.</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

Determination of Urea Adulteration in Milk by Terahertz Spectroscopy

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Hongtao Zhang,
  • Shijie He,
  • Jiahui Gao,
  • Lian Tan,
  • Zhongyang Li

摘要

Milk is a nutritional staple, yet its safety is compromised by the illegal practice of adding urea to falsely inflate protein content, with concentrations exceeding 0.5 g/L posing severe health risks. To overcome the limitations of current complex, time-consuming, and destructive detection methods, we introduce a rapid, non-destructive quantitative approach for urea detection in liquid milk using THz-TDS coupled with chemometrics. Milk-urea samples across a broad range of concentrations, from 0 to 50% (w/w), were prepared to ensure robust calibration, and the corresponding absorption coefficient spectra were collected within the 0.22–1.7 THz range. Optimization of preprocessing involved evaluating seven algorithms; the combined SNV + SG smoothing proved optimal, enhancing the PLS model’s prediction set determination coefficient (R2P) to 0.9408. Subsequently, 16 quantitative regression models were developed by combining four feature selection methods (CARS, SPA, UVE, VISSA) and four regression algorithms (SVR, LSTM, RF, PSO-BP). The VISSA-RF model, which integrates VISSA dimensionality reduction with RF regression, consistently exhibited the superior predictive performance. This optimal model yielded a prediction set R2P of 0.9845 and a root mean square error of prediction (RMSEP) of 0.0197, marking an accuracy increase of up to 1.38% over the 15 alternative methods. Ultimately, this study validates the high-precision and robust capability of THz spectroscopy for the non-destructive quantitative determination of urea adulterants, providing a novel and efficient technical pathway for on-line and rapid screening in dairy product safety monitoring.