Abstract: This study aimed to establish a reliable predictive model for quantitative analysis of gluten in wheat flour using near infrared (NIR) spectroscopy. The optimal spectral pretreatment method combined with simulated annealing algorithm (SAA) was obtained by comparison of the partial least squares (PLS) regression models developed after different spectral pretreatments alone and combined with SAA based on their coefficient of determination (R2), root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP). The results indicated that the stability and prediction performance of the PLS model were greatly improved by using spectral pretreatment combined with SAA, as demonstrated by an increase in R2 from 0.763 7 to 0.949 1, a reduction in RMSEC from 1.371 2 to 0.589 8, and a decrease in RMSEP from 1.450 2 to 0.534 1. The combination of spectral pretreatment and SAA was feasible for the development of a predictive model for quantitative analysis of gluten. Moreover, the optimized model exhibited better stability and prediction performance than the unoptimized model and the one developed with spectral pretreatment alone.

Key words: near infrared spectroscopy, simulated annealing algorithm (SAA), spectral pretreatment, partial least squares (PLS), gluten