Machine learning-assisted hyperspectral reflectance analysis for non-destructive detection of clothianidin residues in tea chrysanthemum

Jingshan Lu, Qiuyan Zhang, Qimo Qi, Gangjun Zheng, Jiuyuan Zhang, Sumei Chen, Fei Zhang, Weimin Fang, Zhiyong Guan, Fadi Chen,
Machine learning-assisted hyperspectral reflectance analysis for non-destructive detection of clothianidin residues in tea chrysanthemum,
Microchemical Journal,
Volume 219,
2025,
115935,
ISSN 0026-265X,
https://doi.org/10.1016/j.microc.2025.115935.
(https://www.sciencedirect.com/science/article/pii/S0026265X25032837)
Abstract: Pesticide residues are critical indicators of tea chrysanthemum quality and safety, yet traditional detection methods are destructive and unsuitable for real-time monitoring. Hyperspectral remote sensing (HRS) offers a rapid, non-destructive alternative, but its application for pesticide residue detection in tea chrysanthemum remains limited. In this study, we conducted a root irrigation experiment with varying concentrations of clothianidin (CLO) and collected leaf hyperspectral reflectance data (400–2400 nm). We evaluated the performance of spectral indices and machine learning algorithms in estimating CLO residues. Residue levels in leaves positively correlated with those in inflorescences, with leaf residues exceeding 1.88 μg/g at 14 days predicting excessive levels at harvest. Sensitive spectral bands were primarily in the green, red-edge, and short-wave infrared regions. Among spectral indices, the difference spectral index (DSI (585, 695)) achieved the highest predictive accuracy (R2 = 0.432, RMSE = 0.502 μg/g). Integrating full-spectrum data with machine learning further improved estimation, with the partial least squares regression (PLSR) model performing best (R2 = 0.654, RMSE = 0.438 μg/g). These findings demonstrate the feasibility of using HRS technology for non-destructive pesticide residue monitoring in tea chrysanthemum. This approach provides a novel, efficient method for ensuring product safety and quality, with potential applications in precision agriculture and food safety monitoring.
Keywords: Hyperspectral; Clothianidin; Pesticide residue; Spectral indices; Machine learning