Development of simplified models for nondestructive testing of rice (with husk) protein content using hyperspectral imaging technology

• Published in: Vibrational spectroscopy 2021-05, Vol.114, p.103230, Article 103230
• Main Authors: Ma, Chengye, Ren, Zhishang, Zhang, Zhehao, Du, Juan, Jin, Chengqian, Yin, Xiang
• Format: Article
• Language: English
• Subjects: Hyperspectral imaging; Multivariate analysis; Optimal wavelength selection; Protein content; Rice; Visualisation
• ISSN: 0924-2031; 1873-3697
• DOI: 10.1016/j.vibspec.2021.103230

•The sample covers 87 different rice (with chaff) varieties in China.•Samples were scanned with a high-energy wavelength range of 938-−2215 nm and the spectral image absorption peak of the sample appears obviously.•The optimal characteristic wavelengthcombination was selected.•According to the visualization image, the distribution of rice protein could be understood. The study aimed to establish predictive models of protein content in rice (with husk) using a hyperspectral imaging (HSI) system in a collection of 87 rice varieties in China in the wavelength range of 938–2215 nm and the first established multivariate calibration models over the full wavelength range by using partial least-square regression (PLSR), principal component regression (PCR) and least-square support vector regression (LS-SVR). In predictive model optimisation, the optimal wavelengths were selected by using regression coefficients (RC) as discriminating factors to establish PLSR models. According to the RC of models, the optimal wavelength combination was 17 and 7 characteristics. The model based on the 17-characteristic wavelengths was determined as the optimal optimisation model with coefficients of determination for prediction of 0.8011 and root mean square error of prediction of 0.52. The mapping of protein content was achieved by transferring a quantitative model to each pixel. According to the visualisation image, the distribution of rice protein could be understood, thus realising the possibility of on-line detection of protein content by using HSI technology.