Scholarlink

<p>Deoxynivalenol (DON), found in corn, is a serious food safety issue. This study utilized a hyperspectral imaging system (HSI) in the shortwave infrared region (reflectance, 900–2500&#xa0;nm) to estimate the deoxynivalenol (DON) content in corn kernels through correlated spectral signatures. The corn kernel pericarp layers were cracked and spiked with laboratory DON at five concentration levels—0, 1, 2, 5, and 10&#xa0;µg/g to mimic the natural distribution of DON. The HSI images were acquired at two different orientations of corn grain—germ-side and endosperm-side. The acquired images were subjected to 15 different preprocessing and feature selection methods. Partial least square regression (PLSR) and support vector machine regression (SVMR) models were developed to correlate the processed spectra with the DON content measured by ELISA. The spectral data from the full spectrum and the spectral data from significant wavelengths obtained using feature selection methods were used to build regression models. The SVMR model developed from the germ-side full spectrum with SNV preprocessing provided the best <i>R</i><sup>2</sup> prediction of 0.9855 and RMSE prediction of 0.2953. The SVMR model developed using germ-side significant wavelengths with orthogonal spectral correction (OSC) + standard normal variate (SNV) preprocessing provided the best <i>R</i><sup>2</sup> prediction of 0.9847 and RMSE prediction of 0.3010.</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

Application of non-destructive and chemical-free short-wave infrared hyperspectral imaging (SWIR-HSI) coupled with machine learning regression for rapid estimation of deoxynivalenol (DON) in individual corn kernels

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Rathna Priya Thangaraj Sundaramurthy,
  • Thiruppathi Senthilkumar,
  • Chandra B. Singh,
  • Annamalai Manickavasagan

摘要

Deoxynivalenol (DON), found in corn, is a serious food safety issue. This study utilized a hyperspectral imaging system (HSI) in the shortwave infrared region (reflectance, 900–2500 nm) to estimate the deoxynivalenol (DON) content in corn kernels through correlated spectral signatures. The corn kernel pericarp layers were cracked and spiked with laboratory DON at five concentration levels—0, 1, 2, 5, and 10 µg/g to mimic the natural distribution of DON. The HSI images were acquired at two different orientations of corn grain—germ-side and endosperm-side. The acquired images were subjected to 15 different preprocessing and feature selection methods. Partial least square regression (PLSR) and support vector machine regression (SVMR) models were developed to correlate the processed spectra with the DON content measured by ELISA. The spectral data from the full spectrum and the spectral data from significant wavelengths obtained using feature selection methods were used to build regression models. The SVMR model developed from the germ-side full spectrum with SNV preprocessing provided the best R2 prediction of 0.9855 and RMSE prediction of 0.2953. The SVMR model developed using germ-side significant wavelengths with orthogonal spectral correction (OSC) + standard normal variate (SNV) preprocessing provided the best R2 prediction of 0.9847 and RMSE prediction of 0.3010.