Scholarlink

<p>This study systematically investigated the influence of different detection distances on the spectral detection models of moisture content and Soluble Solid Content in small white apricots, aiming to provide distance parameter references for the practical application of portable spectrometers. Using 98 samples of small white apricots as the subjects, visible/near-infrared diffuse reflectance spectra were collected at five distances of 0.1, 0.2, 0.3, 0.4, and 0.5&#xa0;m using a self-built adjustable distance platform. During the modeling process, different distances are treated as external variables, and spectral reflectance data at various distances are used as inputs for the predictive model, which is then followed by the introduction of the Partial Least Squares Regression (PLSR) model. After preprocessing the original spectral data and eliminating abnormal samples through Monte Carlo cross-validation, the competitive adaptive re-weighted algorithm was employed to screen characteristic wavelengths. The models for moisture content and soluble solids retained 109 and 89 key variables, respectively (out of the original 601). The results indicate that after applying different optimal preprocessing methods to spectra from various detection distances and using CARS for variable selection, the model performance was significantly improved: the coefficient of determination and RPIQ of the optimal model for moisture content increased from 0.8630 to 3.9785 to 0.9245 and 5.3667, respectively, while the root mean square error of the prediction set decreased from 2.4354 to 2.0157; the evaluation metrics of the optimal model for soluble solids increased from the original values of 0.6845 and 1.8691 to 0.7793 and 2.8745, and the root mean square error of the prediction set decreased from the original value of 0.8734 to 0.6834. The optimal detection distance for moisture content is 0.3&#xa0;m (with moving window smoothing preprocessing), while the optimal detection distance for soluble solids is 0.4&#xa0;m (with multi-scattering correction preprocessing). Furthermore, the preliminarily explored segmented weighted multi-distance spectral fusion strategy can integrate complementary spectral features, further enhancing the robustness of the model. This study confirms that the detection distance significantly affects the accuracy of spectral detection of fruit internal quality, providing methodological references for multi-source information fusion modeling and the on-site application of spectral technology.</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

Research on the spectral detection effect and data fusion of small white apricot quality based on different detection distances

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • Yuesen Tong,
  • Huaping Luo,
  • Xueting Ma,
  • Jinlong Yu,
  • Huaiyu Liu,
  • Lei Kang

摘要

This study systematically investigated the influence of different detection distances on the spectral detection models of moisture content and Soluble Solid Content in small white apricots, aiming to provide distance parameter references for the practical application of portable spectrometers. Using 98 samples of small white apricots as the subjects, visible/near-infrared diffuse reflectance spectra were collected at five distances of 0.1, 0.2, 0.3, 0.4, and 0.5 m using a self-built adjustable distance platform. During the modeling process, different distances are treated as external variables, and spectral reflectance data at various distances are used as inputs for the predictive model, which is then followed by the introduction of the Partial Least Squares Regression (PLSR) model. After preprocessing the original spectral data and eliminating abnormal samples through Monte Carlo cross-validation, the competitive adaptive re-weighted algorithm was employed to screen characteristic wavelengths. The models for moisture content and soluble solids retained 109 and 89 key variables, respectively (out of the original 601). The results indicate that after applying different optimal preprocessing methods to spectra from various detection distances and using CARS for variable selection, the model performance was significantly improved: the coefficient of determination and RPIQ of the optimal model for moisture content increased from 0.8630 to 3.9785 to 0.9245 and 5.3667, respectively, while the root mean square error of the prediction set decreased from 2.4354 to 2.0157; the evaluation metrics of the optimal model for soluble solids increased from the original values of 0.6845 and 1.8691 to 0.7793 and 2.8745, and the root mean square error of the prediction set decreased from the original value of 0.8734 to 0.6834. The optimal detection distance for moisture content is 0.3 m (with moving window smoothing preprocessing), while the optimal detection distance for soluble solids is 0.4 m (with multi-scattering correction preprocessing). Furthermore, the preliminarily explored segmented weighted multi-distance spectral fusion strategy can integrate complementary spectral features, further enhancing the robustness of the model. This study confirms that the detection distance significantly affects the accuracy of spectral detection of fruit internal quality, providing methodological references for multi-source information fusion modeling and the on-site application of spectral technology.