Determination of glyphosate residues in lentils using near-infrared hyperspectral imaging coupled with chemometric regression techniques

• Near-infrared hyperspectral imaging (HSI) effectively detects glyphosate residues in lentil samples. • Different lentil varieties exhibit unique spectral characteristics influenced by glyphosate concentration. • Partial least squares regression (PLSR) with orthogonal signal correction (OSC) achieves high prediction accuracy in glyphosate detection. • VIP and selectivity ratio (sRatio) methods improve model accuracy by optimizing wavelength selection. • The proposed HSI-PLSR model enables rapid, non-destructive glyphosate screening in lentils, enhancing food safety monitoring. The accurate measurement of pesticide content in lentils throughout the supply chain is essential to ensure compliance with the maximum residue level (MRL) regulations set by government agencies. The objective of this research was to study the feasibility of using near-infrared (NIR) hyperspectral imaging (HSI) system in the 900–2500 nm wavelength range to detect glyphosate residue levels in black beluga lentil, red lentil, large green lentil, and French green lentil at five glyphosate concentration levels (0 (control), 5, 10, 15 and 20 mg/kg). The prediction of the glyphosate content was achieved by developing partial least squares regression (PLSR) and principal component regression (PCR) models using different spectral preprocessing techniques on full spectrum and variables selected by selectivity ratio (sRatio) and variable importance in projection (VIP) methods. The full spectrum results showed that in black beluga lentil orthogonal spectral correction (OSC)-PLSR dataset performed best with correlation coefficient of prediction (R 2 p ), root mean square error of prediction (RMSEP) and residual predictive deviation (RPD) values of 0.916, 2.294 and 3.805, respectively. Further in red lentil the standard normal variate (SNV)-1 st derivative-PLSR performed best with a R 2 p , RMSEP and RPD values were 0.920, 2.190 and 3.925, respectively. Whereas in large green lentil and French green lentil, the 1st derivative-PLSR performed efficient with a R 2 p , RMSEP and RPD values were 0.938, 1.900 and 3.623 and 0.929, 2.017 and 3.413, respectively. Further, based on the wavelengths selected by VIP method, the OSC-VIP-PLSR model performed best for black beluga lentil with R 2 p , RMSEP and RPD values of 0.933, 1.915 and 3.595, respectively. In red lentil, the VIP-SNV + 1st derivative-PLSR showed highest performance with R 2 p , RMSEP and RPD values of 0.925, 2.066 and 3.332, respectively. Whereas, in large green lentil and French green lentil VIP-1st derivative-PLSR depicted highest prediction accuracy with R 2 p , RMSEP and RPD 0.940, 1.741 and 3.954 and 0.941, 1.726 and 3.988, respectively. This study demonstrated that the hyperspectral imaging system in 900–2500 nm range combined with machine learning could be used for the rapid and accurate glyphosate detection in lentils.