Combined use of near infrared spectroscopy and chemometrics for the simultaneous detection of multiple illicit additions in wheat flour

• Multiple illegal additives level in flour was simultaneously detected using NIRS. • LSTM dimensionality reduction was combined with PLS to build quantitative model. • LSTM network parameter was optimized by Bayesian algorithm to improve performance. The safety and quality of wheat flour, a staple in daily life, directly affect people's health. The illegal use of additives such as azodicarbonamide, talcum and gypsum powders can lead to serious health risks. Traditional detection methods are time-consuming and unsuitable for routine screening, creating an urgent need for rapid, non-destructive techniques. To identify multiple illicit additives within wheat flour, a simultaneous detection model for three additives—azodicarbonamide, talcum and gypsum powders—was constructed using near-infrared spectroscopy and chemometrics methods. The model combines long short-term memory network (LSTM) data dimensionality reduction with partial least squares to detect multiple illicit additives in wheat flour. The Bayesian optimization algorithm was used to optimize the LSTM parameters. Compared to regression models built with competitive adaptive reweighted sampling and genetic algorithm for feature wavelength selection, the performance improved significantly, enhancing generalization capability. The validation set’s coefficients of determination were 0.9828, 0.9771, 0.9765, with root mean square errors of 0.0008 %, 0.2915 %, 0.2822 %, and residual prediction errors of 7.4067, 6.4020, and 6.2159, respectively. Combining near-infrared spectroscopy with LSTM dimensionality reduction and partial least squares enables quick, simultaneous detection of various illicit additives in wheat flour, providing a novel approach for efficient and precise inspection.