Scholarlink

<p>This study investigates the potential of laser light backscattering imaging (LLBI) at five wavelengths (532, 650, 780, 880, and 980&#xa0;nm) combined with different machine learning (ML) algorithms for classification of ripening stages and prediction of key quality attributes including flesh firmness (FF), soluble solids content (SSC), and titratable acidity (TA) during postharvest storage. Principal component analysis (PCA) and correlation-based feature selection (CFS) were applied to identify the most informative image descriptors and reduce model complexity. LLBI features at 650&#xa0;nm and 780&#xa0;nm provided the most reliable results. The most accurate classification of kiwifruits according to ripening stage was obtained by multilayer perceptron neural networks (MLPNN) using CFS-selected features at 780&#xa0;nm, with accuracy of 94.00% in test stage. Strong relationships were observed between LLBI features and physicochemical attributes. The FF, SSC, and TA of kiwifruits were successfully predicted by LLBI (650&#xa0;nm)-CFS-MLPNN model with coefficient of determination (R<sup>2</sup>) of 0.9557, 0.9115, and 0.9268, in test stage, respectively. These results indicate that the proposed approach provides a reliable laboratory-scale solution for non-destructive monitoring of kiwifruit ripeness during postharvest storage. However, further research and system-level optimization are required to validate its robustness and feasibility for large-scale, on-line industrial applications aimed at reducing waste and improving operational efficiency.</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

Combining multi-wavelength laser backscattering imaging and machine learning techniques for non-destructive assessment of kiwifruit ripeness

  • 期刊论文

  • 发表日期：

    2026年4月14日

查看全文
  • Mehran Azizi,
  • Hemad Zareiforoush,
  • Adel Bakhshipour

摘要

This study investigates the potential of laser light backscattering imaging (LLBI) at five wavelengths (532, 650, 780, 880, and 980 nm) combined with different machine learning (ML) algorithms for classification of ripening stages and prediction of key quality attributes including flesh firmness (FF), soluble solids content (SSC), and titratable acidity (TA) during postharvest storage. Principal component analysis (PCA) and correlation-based feature selection (CFS) were applied to identify the most informative image descriptors and reduce model complexity. LLBI features at 650 nm and 780 nm provided the most reliable results. The most accurate classification of kiwifruits according to ripening stage was obtained by multilayer perceptron neural networks (MLPNN) using CFS-selected features at 780 nm, with accuracy of 94.00% in test stage. Strong relationships were observed between LLBI features and physicochemical attributes. The FF, SSC, and TA of kiwifruits were successfully predicted by LLBI (650 nm)-CFS-MLPNN model with coefficient of determination (R2) of 0.9557, 0.9115, and 0.9268, in test stage, respectively. These results indicate that the proposed approach provides a reliable laboratory-scale solution for non-destructive monitoring of kiwifruit ripeness during postharvest storage. However, further research and system-level optimization are required to validate its robustness and feasibility for large-scale, on-line industrial applications aimed at reducing waste and improving operational efficiency.