Non-invasive detection of protein content in corn distillers dried grains with solubles: method for selecting spectral variables to construct high-performance calibration model using near infrared reflectance spectroscopy

Corn distillers dried grains with solubles (DOGS), a byproduct of the bioethanol industry, is commonly used as animal feed. This paper evaluates the use of backward variable selection partial least square (BVSPLS) and genetic algorithm (GA) methods to select the spectral variables of near infrared (NIR) reflectance spectroscopy and construct high-performance calibration models of protein content in corn DDGS. The BVSPLS analysis utilised 16% of the spectral variables. Compared to the full spectrum model, the model constructed from the variables selected by the BVSPLS analysis significantly improved the accuracy of the model fit and achieved a 19% decrease in the standard error of validation (SEP) and a 23% increase in the residual validation deviation (RPD). The GA analysis selected 8% of the total NIR spectral variables and the model constructed from these selected variables had a fitted accuracy comparable to that of the full spectrum model. The spectral variables selected by both the BVSPLS analysis and GA analysis significantly simplified the NIR calibration model and provided better correlation between the selected spectral variables and protein content of corn DOGS. These results also have important implications for the development of a rapid, non-invasive, online analysis system to detect protein content of corn DDGS in-situ.