Analyzing protein concentration from intact wheat caryopsis using hyperspectral reflectance

Abstract

Abstract Winter wheat grain samples from 185 sites across southern Shanxi region were processed and analyzed by using a non-destructive approach. For this purpose, spectral data and protein content data of grain and grain powder were obtained. After combining six types of pre-processed spectra and four types of multivariate statistical models, a relationship hyperspectral datasets and grain protein content is presented. It was found that the hyperspectral reflectance of winter wheat grain and powder was positively correlated with the protein contents, which provide the possibility for hyperspectral quantitative assessment. The spectral characteristic bands of protein content in winter wheat extracted based on the SPA algorithm were proved to be around 350–430 nm; 851–1154 nm; 1300–1476 nm; and 1990–2050 nm. In powder samples, SG-BPNN had the best monitoring effect, with the accuracy of Rv2 = 0.814, RMSEv = 0.024, and RPDv = 2.318. While in case of grain samples, the SG-SVM model exhibited the best monitoring effect, with the accuracy of Rv2 = 0.789, RMSEv = 0.026, and RPDv = 2.177. Based on the experimental findings, we propose that a combination of spectral pretreatment and multivariate statistical modeling is helpful for the non-destructive and rapid estimation of protein content in winter wheat.