Quantitative determination of multiple illegal additives in flour based on near-infrared spectroscopy

Accurate quantification of illegal additives in flour is critical for ensuring food safety. Conventional detection methods, however, suffer from low efficiency and are inadequate for rapid quantitative analysis. Herein, we presented a near-infrared spectroscopy-based quantitative detection approach for the simultaneous determination of Azodicarbonamide (ADA), Talcum Powder (TalcP), and Gypsum Powder (GypsumP) in flour. Three wavelength selection algorithms were respectively combined with three multivariate regression methodologies to construct rapid detection model of three illegal additives in flour. The genetic simulated annealing algorithm was applied for modeling parameter optimization in the nonlinear modeling process. Results showed that the combination of competitive adaptive reweighted sampling and kernel extreme learning machine achieved the best detection performance for ADA, TalcP, and GypsumP. The prediction set’s coefficients of determination were 0.9855, 0.9917, and 0.9875; the root mean square errors were 0.0008, 0.1889, and 0.2354; and the residual prediction deviations were 8.2928, 11.0041, and 8.9388, respectively. The detection accuracy meets the requirements of practical applications. This approach provides a scientific basis for flour quality detection and strong technical support for ensuring food safety and consumer health.