Scholarlink

<p>Crude protein content is a key quality parameter in cereal processing, yet conventional reference methods such as Dumas combustion are destructive and require sample preparation. Hyperspectral imaging (HSI) extends conventional point spectroscopy by coupling broadband spectral acquisition with camera-based imaging. This study presents the design, assembly, and characterization of a custom-built two-dimensional VIS/NIR HSI system continuously covering 400–1600&#xa0;nm within a single measurement configuration. The system integrates illumination-side wavelength selection using a Czerny–Turner monochromator, synchronized VIS and NIR area-scan cameras, a xenon arc lamp, and control electronics. The broadband range was deliberately selected to capture scattering and autofluorescence-related contributions in the visible range alongside chemically specific overtone absorption bands in the near-infrared, providing complementary information. As a proof of concept, a feedforward neural network (FFNN; N–25–1) was trained and evaluated under grouped nested cross-validation (outer 5 × 5; inner threefold) on 27 subsample groups derived from six starch–gluten powder mixtures (reference: Dumas combustion). The model achieved an R<sup>2</sup> of 0.86 ± 0.15 and an RMSE of 0.97 ± 0.54% (w/w) for crude protein estimation across the outer test folds. These results demonstrate the feasibility of non-destructive VIS–NIR hyperspectral imaging for crude protein determination in cereal powder matrices. Future work will extend validation to real wheat flour samples, incorporate additional quality parameters, and investigate systematic wavelength selection to enable compact, cost-effective multispectral sensor concepts.</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

Crude protein prediction in grain-based powder systems using VIS–NIR hyperspectral imaging and feedforward neural networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Ronny Takacs,
  • Günther Gaßner,
  • Dominik Geier,
  • Thomas Becker

摘要

Crude protein content is a key quality parameter in cereal processing, yet conventional reference methods such as Dumas combustion are destructive and require sample preparation. Hyperspectral imaging (HSI) extends conventional point spectroscopy by coupling broadband spectral acquisition with camera-based imaging. This study presents the design, assembly, and characterization of a custom-built two-dimensional VIS/NIR HSI system continuously covering 400–1600 nm within a single measurement configuration. The system integrates illumination-side wavelength selection using a Czerny–Turner monochromator, synchronized VIS and NIR area-scan cameras, a xenon arc lamp, and control electronics. The broadband range was deliberately selected to capture scattering and autofluorescence-related contributions in the visible range alongside chemically specific overtone absorption bands in the near-infrared, providing complementary information. As a proof of concept, a feedforward neural network (FFNN; N–25–1) was trained and evaluated under grouped nested cross-validation (outer 5 × 5; inner threefold) on 27 subsample groups derived from six starch–gluten powder mixtures (reference: Dumas combustion). The model achieved an R2 of 0.86 ± 0.15 and an RMSE of 0.97 ± 0.54% (w/w) for crude protein estimation across the outer test folds. These results demonstrate the feasibility of non-destructive VIS–NIR hyperspectral imaging for crude protein determination in cereal powder matrices. Future work will extend validation to real wheat flour samples, incorporate additional quality parameters, and investigate systematic wavelength selection to enable compact, cost-effective multispectral sensor concepts.