Thai fragrant rice, renowned worldwide for its distinctive texture and unique jasmine aroma, is one of the most widely exported rice varieties globally. However, the market has seen the emergence of ordinary, low-cost rice with similar appearance being fraudulently sold as Thai fragrant rice, or mixed into genuine Thai fragrant rice, making it difficult for consumers to distinguish authenticity by visual inspection alone. This situation seriously undermines consumers’ legitimate rights and interests. This study addresses the issue of adulteration in the Thai fragrant rice market, where conventional rice or flavored rice with added aroma is used as fraudulent substitutes. It proposes a rapid, non-destructive quantitative identification method based on the integration of fluorescence hyperspectral imaging technology and machine learning techniques. This study focused on Thai fragrant rice and Anhui rice adulterated with Thai flavoring, employing fluorescence hyperspectral technology combined with machine learning to quantitatively identify adulteration in Thai fragrant rice. Fluorescence hyperspectral data were preprocessed using Multiplicative Scatter Correction (MSC), Min–Max Normalization (MMN), and Standard Normal Variate (SNV). Feature wavelength selection was performed using Bootstrapping Soft Shrinkage (BOSS), Competitive Adaptive Reweighted Sampling (CARS), and Successive Projections Algorithm (SPA). The results demonstrated that, after preprocessing and feature selection, quantitative detection of Thai fragrant rice adulteration could be achieved using Support Vector Regression (SVR), Partial Least Squares Regression (PLSR), and Least Squares Support Vector Regression (LSSVR). Finally, the quantitative identification models were optimized using the Archimedes Optimization Algorithm (AOA) and Snake Optimizer (SO). The best-performing model, SNV-BOSS-LSSVR-AOA, achieved an RMSEC of 0.0014, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({R}_{C}^{2}\)</EquationSource> </InlineEquation> of 0.9999, RMSEP of 0.0327, and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({R}_{P}^{2}\)</EquationSource> </InlineEquation> of 0.9893, successfully accomplishing quantitative identification of Thai fragrant rice adulteration. This study confirms the reliability of fluorescence hyperspectral technology combined with optimized machine learning models for quantitative identification of Thai jasmine rice adulteration, enabling rapid detection of Thai jasmine rice adulteration issues in the market, ensuring the authenticity of Thai jasmine rice sales, and standardizing Thai jasmine rice market regulations.

Research on non-destructive quantitative detection of adulterated thai rice using fluorescence hyperspectral technology and machine learning

期刊论文
发表日期：

2026年3月6日

查看全文

Rui Qing,
Rongsheng Fan,
Kunyu Li,
Lei Wang,
Wenliang Zhang,
Tieen Xia,
Zhiliang Kang

摘要

Thai fragrant rice, renowned worldwide for its distinctive texture and unique jasmine aroma, is one of the most widely exported rice varieties globally. However, the market has seen the emergence of ordinary, low-cost rice with similar appearance being fraudulently sold as Thai fragrant rice, or mixed into genuine Thai fragrant rice, making it difficult for consumers to distinguish authenticity by visual inspection alone. This situation seriously undermines consumers’ legitimate rights and interests. This study addresses the issue of adulteration in the Thai fragrant rice market, where conventional rice or flavored rice with added aroma is used as fraudulent substitutes. It proposes a rapid, non-destructive quantitative identification method based on the integration of fluorescence hyperspectral imaging technology and machine learning techniques. This study focused on Thai fragrant rice and Anhui rice adulterated with Thai flavoring, employing fluorescence hyperspectral technology combined with machine learning to quantitatively identify adulteration in Thai fragrant rice. Fluorescence hyperspectral data were preprocessed using Multiplicative Scatter Correction (MSC), Min–Max Normalization (MMN), and Standard Normal Variate (SNV). Feature wavelength selection was performed using Bootstrapping Soft Shrinkage (BOSS), Competitive Adaptive Reweighted Sampling (CARS), and Successive Projections Algorithm (SPA). The results demonstrated that, after preprocessing and feature selection, quantitative detection of Thai fragrant rice adulteration could be achieved using Support Vector Regression (SVR), Partial Least Squares Regression (PLSR), and Least Squares Support Vector Regression (LSSVR). Finally, the quantitative identification models were optimized using the Archimedes Optimization Algorithm (AOA) and Snake Optimizer (SO). The best-performing model, SNV-BOSS-LSSVR-AOA, achieved an RMSEC of 0.0014, \({R}_{C}^{2}\) of 0.9999, RMSEP of 0.0327, and \({R}_{P}^{2}\) of 0.9893, successfully accomplishing quantitative identification of Thai fragrant rice adulteration. This study confirms the reliability of fluorescence hyperspectral technology combined with optimized machine learning models for quantitative identification of Thai jasmine rice adulteration, enabling rapid detection of Thai jasmine rice adulteration issues in the market, ensuring the authenticity of Thai jasmine rice sales, and standardizing Thai jasmine rice market regulations.