Scholarlink

<p>The primary objective of this study is to demonstrate the feasibility of employing Laser-Induced Breakdown Spectroscopy (LIBS) in combination with machine learning algorithms for the identification and assessment of abiotic stress in plants. The machine learning approaches considered are multilinear regression (MLR), support vector regression (SVR), partial least squares regression (PLSR), least absolute shrinkage and selection operator (LASSO), and Gaussian process regression (GPR). The stress condition (based on nutrient content - Ca and K) in the sample is also estimated using the calibration-free LIBS (CF-LIBS) method. The experiments are carried out by varying the laser irradiances and stand-off plasma collection distances. It is observed that the Ca concentration in the normal sample is higher than the K concentration, regardless of the incident laser irradiance and stand-off plasma collection distance. The opposite trend is observed in abiotically stressed samples. A clear distinction is observed between normal and abiotically stressed samples in the LIBS spectrum. The GPR method trained on normalized dataset (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\dfrac{I_{\lambda }}{I_{656 nm}}\)</EquationSource> </InlineEquation>) resulted in achieving better estimate of Ca and K in normal (R<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(^{\varvec{2}}\)</EquationSource> </InlineEquation> = 0.94, Ca = 47891 <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\varvec{\pm }\)</EquationSource> </InlineEquation> 1895 ppm; R<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(^{\varvec{2}}\)</EquationSource> </InlineEquation> = 0.90, K= 21234 <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\varvec{\pm }\)</EquationSource> </InlineEquation>2147 ppm) and abiotic stressed samples (R<InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(^{\varvec{2}}\)</EquationSource> </InlineEquation> = 0.92, Ca = 38715 <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(\varvec{\pm }\)</EquationSource> </InlineEquation> 1985 ppm; R<InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(^{\varvec{2}}\)</EquationSource> </InlineEquation> = 0.91, K = 57857 <InlineEquation ID="IEq9"> <EquationSource Format="TEX">\(\varvec{\pm }\)</EquationSource> </InlineEquation> 2458 ppm). The CF-LIBS based Ca and K estimate (SoD= 0.5 m, irradiance = 2.5x10<InlineEquation ID="IEq10"> <EquationSource Format="TEX">\(^{\varvec{10}}\)</EquationSource> </InlineEquation> W/cm<InlineEquation ID="IEq11"> <EquationSource Format="TEX">\(^{\varvec{2}}\)</EquationSource> </InlineEquation>) is in agreement with GPR method trained on normalised dataset, and atomic emission spectroscopy (AES) measurements. The normalized LIBS dataset yielded zero misclassification when evaluated using the Random Forest classifier. The study concluded that the normalized LIBS–GPR model can be adapted for real-time operation following calibration of SoD and laser irradiance.</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

Identification of abiotic stress in Psidium guajava plant sample using LIBS method combined with machine learning approach

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Veerappan Kaliyaperumal,
  • Sathiesh Kumar Vajravelu,
  • Aiswarya Justin,
  • Thangaraja Maruthaiah

摘要

The primary objective of this study is to demonstrate the feasibility of employing Laser-Induced Breakdown Spectroscopy (LIBS) in combination with machine learning algorithms for the identification and assessment of abiotic stress in plants. The machine learning approaches considered are multilinear regression (MLR), support vector regression (SVR), partial least squares regression (PLSR), least absolute shrinkage and selection operator (LASSO), and Gaussian process regression (GPR). The stress condition (based on nutrient content - Ca and K) in the sample is also estimated using the calibration-free LIBS (CF-LIBS) method. The experiments are carried out by varying the laser irradiances and stand-off plasma collection distances. It is observed that the Ca concentration in the normal sample is higher than the K concentration, regardless of the incident laser irradiance and stand-off plasma collection distance. The opposite trend is observed in abiotically stressed samples. A clear distinction is observed between normal and abiotically stressed samples in the LIBS spectrum. The GPR method trained on normalized dataset ( \(\dfrac{I_{\lambda }}{I_{656 nm}}\) ) resulted in achieving better estimate of Ca and K in normal (R \(^{\varvec{2}}\) = 0.94, Ca = 47891 \(\varvec{\pm }\) 1895 ppm; R \(^{\varvec{2}}\) = 0.90, K= 21234 \(\varvec{\pm }\) 2147 ppm) and abiotic stressed samples (R \(^{\varvec{2}}\) = 0.92, Ca = 38715 \(\varvec{\pm }\) 1985 ppm; R \(^{\varvec{2}}\) = 0.91, K = 57857 \(\varvec{\pm }\) 2458 ppm). The CF-LIBS based Ca and K estimate (SoD= 0.5 m, irradiance = 2.5x10 \(^{\varvec{10}}\) W/cm \(^{\varvec{2}}\) ) is in agreement with GPR method trained on normalised dataset, and atomic emission spectroscopy (AES) measurements. The normalized LIBS dataset yielded zero misclassification when evaluated using the Random Forest classifier. The study concluded that the normalized LIBS–GPR model can be adapted for real-time operation following calibration of SoD and laser irradiance.