Scholarlink

<p>Blueberries are rich in anthocyanins and vitamins but highly susceptible to surface defects such as desiccation, bruising, and decay, which compromise quality and safety. This study proposes an integrated hyperspectral reflectance imaging and intelligent algorithm approach for automated detection and classification of blueberry surface defects. Vis-NIR images (398–1016&#xa0;nm) were analyzed using PCA combined with band ratio imaging (W<sub>691.44/819.31</sub>), and a two-dimensional Otsu threshold algorithm achieved 94.0% defect detection accuracy. Key wavelengths were extracted via CARS, SPA, and UVE methods, and classification models—including PLSR, BPNN, and LS-SVM—were developed to categorize blueberries as normal, dried, bruised, or decayed. The UVE-PLSR model achieved the highest accuracy (98.5%) with 119 wavelengths, while SPA-based models required only 5.68% of features yet maintained 97% accuracy, demonstrating a favorable balance between precision and efficiency. These results highlight the method’s robustness and practicality for large-scale industrial inspection, providing a valuable framework for enhancing quality control and reducing post-harvest losses in blueberry production.</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

Hyperspectral reflectance imaging for simultaneous detection and classification of surface defects in blueberries

  • 期刊论文

  • 发表日期：

    2025年11月18日

查看全文
  • Xiong Li,
  • Yawen Guo,
  • Xinlin Xiong,
  • Yande Liu,
  • Ouyang Aiguo,
  • Xiangguo He

摘要

Blueberries are rich in anthocyanins and vitamins but highly susceptible to surface defects such as desiccation, bruising, and decay, which compromise quality and safety. This study proposes an integrated hyperspectral reflectance imaging and intelligent algorithm approach for automated detection and classification of blueberry surface defects. Vis-NIR images (398–1016 nm) were analyzed using PCA combined with band ratio imaging (W691.44/819.31), and a two-dimensional Otsu threshold algorithm achieved 94.0% defect detection accuracy. Key wavelengths were extracted via CARS, SPA, and UVE methods, and classification models—including PLSR, BPNN, and LS-SVM—were developed to categorize blueberries as normal, dried, bruised, or decayed. The UVE-PLSR model achieved the highest accuracy (98.5%) with 119 wavelengths, while SPA-based models required only 5.68% of features yet maintained 97% accuracy, demonstrating a favorable balance between precision and efficiency. These results highlight the method’s robustness and practicality for large-scale industrial inspection, providing a valuable framework for enhancing quality control and reducing post-harvest losses in blueberry production.