Scholarlink

<p>Accurate identification of dill (<i>Anethum graveolens</i> L.) seed varieties is essential for ensuring quality control in food processing and seed distribution. This study investigated and compared the potential of RGB digital imaging and short-wave infrared (SWIR) hyperspectral imaging (HSI) integrated with machine learning, for the non-destructive classification of three commercially important dill (<i>Anethum graveolens</i> L.) seed varieties: Bouquet, Dukat, and Diana. Digital images were acquired under controlled LED and halogen illumination conditions, and image features were extracted for classification using Random Forest (RF) model. The SWIR HSI system was used to collect reflectance spectra across the 900–2500&#xa0;nm wavelength range to capture detailed spectral responses of the seeds, and machine-learning-based spectral models were developed using Partial Least Squares Discriminant Analysis (PLS-DA), RF, and PLS-DA with Variable Importance in Projection (VIP)-based wavelength selection. Among the digital imaging methods, halogen lighting combined with feature selection slightly outperformed LED-based imaging, achieving a classification accuracy of 96.97%. However, HSI outperformed both, with the PLS-DA model achieving 100% classification accuracy. The VIP analysis identified three effective wavelengths (920.2, 1110.8, and 1263.3&#xa0;nm), enabling a reduced PLS-DA-VIP model that retained the perfect classification performance. These results suggest that while hyperspectral imaging provides superior accuracy, digital imaging still demonstrates strong classification performance and holds potential for use in resource-constrained settings. This study provides a valuable guide for selecting the appropriate technology for dill seed classification based on performance needs and operational scale.</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

Comparative study of digital and hyperspectral imaging for varietal discrimination of dill (Anethum graveolens L.) seeds

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Umesh C. Lohani,
  • L. G. Divyanth,
  • Diksha Singla,
  • T. Senthilkumar,
  • Chandra B. Singh

摘要

Accurate identification of dill (Anethum graveolens L.) seed varieties is essential for ensuring quality control in food processing and seed distribution. This study investigated and compared the potential of RGB digital imaging and short-wave infrared (SWIR) hyperspectral imaging (HSI) integrated with machine learning, for the non-destructive classification of three commercially important dill (Anethum graveolens L.) seed varieties: Bouquet, Dukat, and Diana. Digital images were acquired under controlled LED and halogen illumination conditions, and image features were extracted for classification using Random Forest (RF) model. The SWIR HSI system was used to collect reflectance spectra across the 900–2500 nm wavelength range to capture detailed spectral responses of the seeds, and machine-learning-based spectral models were developed using Partial Least Squares Discriminant Analysis (PLS-DA), RF, and PLS-DA with Variable Importance in Projection (VIP)-based wavelength selection. Among the digital imaging methods, halogen lighting combined with feature selection slightly outperformed LED-based imaging, achieving a classification accuracy of 96.97%. However, HSI outperformed both, with the PLS-DA model achieving 100% classification accuracy. The VIP analysis identified three effective wavelengths (920.2, 1110.8, and 1263.3 nm), enabling a reduced PLS-DA-VIP model that retained the perfect classification performance. These results suggest that while hyperspectral imaging provides superior accuracy, digital imaging still demonstrates strong classification performance and holds potential for use in resource-constrained settings. This study provides a valuable guide for selecting the appropriate technology for dill seed classification based on performance needs and operational scale.