NIR hyperspectral imaging with multivariate analysis for measurement of oil and protein contents in peanut varieties

Abstract

The potential of hyperspectral imaging in the spectral range of 1000–2500 nm with multivariate analysis for the prediction of oil and protein concentrations in five peanut cultivars was investigated. Quantitative partial least squares regression (PLSR) models were established using the extracted spectral data from hyperspectral images and the reference measured oil and protein concentrations. The PLSR models established using the whole spectral data pretreated by the multiplicative scatter correction (MSC) method showed good results for predicting the oil concentration with a determination coefficient (R_P²) of 0.945 and root mean square errors of prediction (RMSEP) of 0.196 and for predicting the protein concentration with R_P² of 0.901 and RMSEP of 0.441. In addition, eight optimal wavelengths were selected for protein and oil contents, respectively, using the regression coefficients of the PLSR analysis and used for simplifying the obtained models. The simplified PLSR models also showed good performances with R_p² of 0.933 and 0.912 for predicting oil and protein concentrations. The whole results demonstrated that the NIR hyperspectral imaging technique coupled with chemometric analysis is a promising tool for rapid and non-destructive determination of oil and protein concentrations in peanut kernels and has the potential to develop a multispectral imaging system for future on-line detection of peanut quality.