Research on rapid non-destructive detection of tannin and protein content in sorghum based on multi-output Gaussian process

Abstract

Sorghum grains' tannin and protein content significantly affect their quality and uses. To address the time-consuming and high-cost issues associated with conventional techniques, this study utilizes near-infrared spectroscopy technology combined with chemometric methods to achieve rapid and non-destructive detection of tannin and protein content in sorghum. Sorghum grains with tannin content of 0.47 % −3.21 % and protein content of 5.37 % −8.23 % were taken as samples, data preprocessing was performed using Standard Normal Variate Transformation (SNV), and the sample set was partitioned using the sample set partitioning based on joint X-Y distances (SPXY) algorithm. A Multi-Output Gaussian Process Regression (MOGPR) model was established and compared with models such as Single-Output Gaussian Process Regression (SOGPR), Partial Least Squares Regression (PLSR), and Back Propagation Neural Network (BPNN). The results indicate that the Gaussian Process Regression (GPR) model significantly outperforms the PLSR and BPNN models, and the MOGPR model clearly surpasses the SOGPR model. In the prediction sets for tannin and protein, R_P² values are 0.9790 and 0.9500, RMSEP values are 0.0587 and 0.1699, and RPD values are 6.8928 and 4.4710, respectively. The results indicate that the MOGPR model established using near-infrared spectroscopy in conjunction with chemometrics is more suitable for the rapid and non-destructive detection of tannins and proteins in sorghum.