Rapid and visual measurement of fat content in peanuts by using the hyperspectral imaging technique with chemometrics

Abstract

Out of all fundamental nutrients in peanuts, the amount of fat is the largest. Fat content is regarded as an important factor that significantly affects the processing of peanuts into different products. In this study, the feasibility of hyperspectral imaging (HSI) for rapidly and non-destructively detecting fat content in peanuts is investigated. An appropriate method was adopted to extract spectral information from the hyperspectral images (900–1700 nm) of different peanut varieties. Based on the extracted spectral information and the corresponding chemical values of fat, the best pre-processing and modeling method was established by comparing different methods. For pretreatment, the methods included standard normal variate (SNV), derivative (der), detrend, etc. For modeling, they included multiple linear regression (MLR), principal component regression (PCR) and partial least squares (PLS). The 2^nd-der-SNV-PLS model generated the best results with a regression coefficient and standard error squares of 0.95 and 0.99 in calibration and of 0.90 and 1.47 in prediction, respectively. A simplified 2^nd-der-SNV-RC-PLS model was established using only twelve optimal wavelengths identified by the regression coefficient (RC). The results showed that the model had a high R_P of 0.84 and a low SEP of 1.88. An image processing algorithm according to the 2^nd-der-SNV-RC-PLS model was then utilized in transforming each pixel into hyperspectral images to obtain fat distribution maps. The results of rapid and non-destructive detection of fat content could be potentially used to visualize the distribution of fat content in peanuts.