Scholarlink

<p>To address the limitations of complex and time-consuming traditional laboratory methods for rice quality assessment, this study employed hyperspectral imaging (920–1700&#xa0;nm) to predict five key physicochemical indicators (moisture, protein, amylose, fatty acid value, alkali spreading value) of 210 rice samples from Jiangsu Province. After comparing spectral preprocessing, feature selection, and machine learning algorithms, an innovative three-level fusion framework was adopted to optimize models, achieving synchronized optimization prediction and quality grading across multiple indicators. The results show that the prediction accuracy for moisture, protein, and amylose is high, with the determination coefficients for validation sets (<i>R²</i><sub><i>v</i></sub>) above 0.96 and residual predictive deviation (<i>RPD</i>) above 5.0. In contrast, the prediction performance for fatty acid value and alkali spreading value is moderate, with <i>RPD</i> values of 1.90 and 2.50, respectively. Additionally, the standard normal variate-principal component analysis-linear discriminant analysis (SNV-PCA-LDA) model achieved 93.33% accuracy for japonica rice quality grading. This study establishes a rapid, non-destructive integrated system for rice quality evaluation, providing technical support for efficient quality control.</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

A study on the quantitative prediction of physicochemical indicators and quality grading of rice in Jiangsu Province based on hyperspectral imaging

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Xiaoqi Zhao,
  • Meng Wang,
  • Luxi Xue,
  • Jianwei Chen,
  • Xuan Luo,
  • Fei Shen,
  • Peng Li,
  • Xueming He

摘要

To address the limitations of complex and time-consuming traditional laboratory methods for rice quality assessment, this study employed hyperspectral imaging (920–1700 nm) to predict five key physicochemical indicators (moisture, protein, amylose, fatty acid value, alkali spreading value) of 210 rice samples from Jiangsu Province. After comparing spectral preprocessing, feature selection, and machine learning algorithms, an innovative three-level fusion framework was adopted to optimize models, achieving synchronized optimization prediction and quality grading across multiple indicators. The results show that the prediction accuracy for moisture, protein, and amylose is high, with the determination coefficients for validation sets (R²v) above 0.96 and residual predictive deviation (RPD) above 5.0. In contrast, the prediction performance for fatty acid value and alkali spreading value is moderate, with RPD values of 1.90 and 2.50, respectively. Additionally, the standard normal variate-principal component analysis-linear discriminant analysis (SNV-PCA-LDA) model achieved 93.33% accuracy for japonica rice quality grading. This study establishes a rapid, non-destructive integrated system for rice quality evaluation, providing technical support for efficient quality control.