Comparison of Different Intelligent Spectral Interval Selection Methods for Near-Infrared Spectroscopy-Based Identification of Simulated Flour Adulteration With Multiple Illegal Additives

Background: Flour is a globally consumed staple food, making its quality and safety crucial for public health. However, the limitations of conventional detection methods and the poor suitability of existing spectral interval selection techniques have constrained the accurate classification of flour adulteration. To address these challenges, this study presents a near-infrared spectroscopy-based approach combined with intelligent spectral interval selection for identifying mixed flour adulteration. Methods: Three hybrid intelligent optimization algorithms were developed: genetic simulated annealing (GSA), genetic particle swarm optimization (GPSO), and simulated annealing particle swarm optimization (SAPSO). These algorithms were integrated with interval partial least squares (iPLS) to develop GSA-iPLS, GPSO-iPLS, and SAPSO-iPLS feature interval selection methods. Highly correlated spectral intervals were selected using the cross-validation classification accuracy of partial least squares discriminant analysis (PLSDA) as the objective function. PLSDA and support vector machine (SVM) classification models were constructed using the selected intervals. Results: All three methods effectively reduced spectral dimensionality and enhanced model performance. Among them, the GSA-iPLS-based PLSDA model demonstrated the best linear classification performance on the test set, achieving an accuracy and F1 score of 94.61% on the test set and an accuracy of 92.26% with an F1 score of 92.84% on the independent validation set. In contrast, the GPSO-iPLS-based SVM model exhibited the highest nonlinear classification performance, with an accuracy and F1 score of 100% on both the test and independent validation sets, respectively. Conclusions: The proposed hybrid intelligent spectral interval selection methods significantly enhance the predictive performance and robustness of flour adulteration detection models. These findings provide a valuable technical framework and theoretical basis for assessing flour quality and safety testing.