Scholarlink

<p>To realize rapid and non-destructive detection of lard-palm oil adulteration, this study used Visible-Near Infrared (VNIR, 400–1000&#xa0;nm) and Short-Wave Infrared (SWIR, 900–1700&#xa0;nm) Hyperspectral Imaging (HSI) techniques, along with preprocessing, feature wavelength selection, and machine learning, to build an adulteration ratio quantitative model and achieve visual detection. Lard-palm oil samples with adulteration ratios of 0%-100% (10% intervals) were prepared. After spectral collection, preprocessing and feature selection were performed, followed by model construction using Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Results showed SWIR outperformed VNIR significantly. This superiority is mainly attributed to lipid-related C–H overtone and combination absorptions in the SWIR region, which are more sensitive to compositional differences between lard and palm oil. The optimal model was SVR combined with Normalization (NOR) preprocessing and Iterative Variable Importance and Selection by Stability Analysis (iVISSA) feature selection, achieving R²<sub>P</sub>=0.9933, RMSEP = 2.5791, RRMSE = 0.0516, Bias = 0.0250, RPD = 12.3010 on the test set. For visualization, the established model was further applied in a pixel-wise manner to generate adulteration distribution maps, with colors transitioning from dark blue (0% adulteration) to dark red (100% adulteration). In conclusion, SWIR-HSI with nonlinear algorithms enables rapid, non-destructive, high-precision oil adulteration detection. The NOR-iVISSA-SVR scheme offers a reliable technical path for oil authenticity testing with great practical application potential.</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

Rapid non-destructive detection and visualization of lard adulteration using hyperspectral imaging coupled with machine learning

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Ling Zhang,
  • Xin Wen,
  • Ying Li,
  • Zhiying Su,
  • Ni Lai,
  • Juncheng Chen,
  • Songlin Lai

摘要

To realize rapid and non-destructive detection of lard-palm oil adulteration, this study used Visible-Near Infrared (VNIR, 400–1000 nm) and Short-Wave Infrared (SWIR, 900–1700 nm) Hyperspectral Imaging (HSI) techniques, along with preprocessing, feature wavelength selection, and machine learning, to build an adulteration ratio quantitative model and achieve visual detection. Lard-palm oil samples with adulteration ratios of 0%-100% (10% intervals) were prepared. After spectral collection, preprocessing and feature selection were performed, followed by model construction using Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Results showed SWIR outperformed VNIR significantly. This superiority is mainly attributed to lipid-related C–H overtone and combination absorptions in the SWIR region, which are more sensitive to compositional differences between lard and palm oil. The optimal model was SVR combined with Normalization (NOR) preprocessing and Iterative Variable Importance and Selection by Stability Analysis (iVISSA) feature selection, achieving R²P=0.9933, RMSEP = 2.5791, RRMSE = 0.0516, Bias = 0.0250, RPD = 12.3010 on the test set. For visualization, the established model was further applied in a pixel-wise manner to generate adulteration distribution maps, with colors transitioning from dark blue (0% adulteration) to dark red (100% adulteration). In conclusion, SWIR-HSI with nonlinear algorithms enables rapid, non-destructive, high-precision oil adulteration detection. The NOR-iVISSA-SVR scheme offers a reliable technical path for oil authenticity testing with great practical application potential.