A novel INFO-SVR-based strategy integrated with terahertz metamaterial sensing for detecting trace thiabendazole residues

Excessive thiabendazole residues in agricultural products pose a safety risk. This study proposes a novel approach for trace-level, non-destructive detection of thiabendazole using terahertz metamaterial enhancement and intelligent optimization modeling. In this study, the “L”-shaped THz metamaterial sensor was fabricated to collect transmission spectra (0.5–3.5 THz). A support vector regression (SVR) model, optimized by the weighted mean of vectors (INFO) algorithm for global hyperparameter optimization, was constructed. Dimensionality reduction was performed using Iteratively Retaining Informative Variables (IRIV), Variable Importance Projection (VIP), and Variable Iterative Spatial Shrinkage Approach (VISSA). The results show that the INFO-SVR model outperformed traditional optimization methods. The VISSA-INFO-SVR model showed the best performance, achieving the Limit of Detection ( LOD ) of 0.697 μg/ml. F-tests and t-tests (p > 0.05) further confirmed the statistical reliability of the prediction results. This study integrates the INFO algorithm with terahertz metamaterial spectral modeling, combined with feature extraction techniques, to achieve highly sensitive, non-destructive detection of thiabendazole residues.