Abstract：As a widely used growth regulator, paclobutrazol plays a significant role in ensuring crop yields and improving the quality of agricultural products. However, suppose it is overused and leaves excessive residues. In that case, it will not only pollute the environment but also adhere to the surface of agricultural products, posing a potential threat to consumer health. Therefore, it is urgent to establish an efficient, sensitive, and non-destructive method for the precise detection of paclobutrazol content in agricultural products. In view of the problems of complex pretreatment, long time consumption, and insufficient sensitivity in traditional detection techniques, this paper proposes a non-destructive and highly sensitive detection method for trace paclobutrazol pesticide residues by combining terahertz metamaterial sensors with intelligent optimization modeling algorithms. In this study, an “L”-shaped composite double-peak structure terahertz metamaterial sensor was designed and fabricated. 21 samples of paclobutrazol solutions with different concentrations were prepared. After adding drops, drying, and cooling, the transmission spectra of each concentration in the range of 0.7～3.5 THz were collected, and their spectral characteristics were analyzed. The spectral data were processed by combining pretreatment algorithms such as 1st D, 2nd D, SNV, and MSC, with feature selection algorithms, including CARS, UVE, IRIV, VIP, VISSA. The educational competitive optimization algorithm (ECO) was introduced to optimize the hyperparameters of the support vector regression (SVR) model, thereby constructing the optimal regression model. The results show that the “L”-shaped structure metamaterial sensor proposed in this study has good resonance enhancement ability; In the range of 0.7～3.5 THz, as the concentration of paclobutrazol increases, the terahertz transmission spectrum shows a significant decrease in amplitude and redshift of the resonance peak, showing a good concentration response relationship; The ECO-SVR model has better model effects in the full band than other optimization methods; The 2nd D-VISSA-ECO-SVR model shows the best prediction accuracy and fitting ability, with the model's RP, RMSEP and MAE being 0.974 9, 0.069 0 and 0.054 4; The LOD of the model was calculated to be 0.215 μg·mL-1 by fitting the real concentration and predicted concentration of paclobutrazol. This paper combines the ECO algorithm with terahertz metamaterial technology. It combines pretreatment and feature extraction techniques to achieve high-sensitivity and non-destructive detection of paclobutrazol residues, verifying the effectiveness and superiority of the ECO optimization algorithm in the field of spectral detection, and providing an efficient and practical technical path for pesticide residue analysis.
毛孝冬，黄志凯，刘燕德，胡 军. 太赫兹超材料谐振增强结合ECO-SVR的痕量多效唑残留高灵敏检测方法研究[J]. 光谱学与光谱分析, 2025, 45(11): 3081-3089.
MAO Xiao-dong, HUANG Zhi-kai, LIU Yan-de, HU Jun. Highly Sensitive Detection Method for Trace Paclobutrazol Residues Based on Terahertz Metamaterial Resonant Enhancement Combined With
ECO-SVR. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(11): 3081-3089.
