A green and efficient method for detecting nicosulfuron residues in field maize using hyperspectral imaging and deep learning.

Accurate and rapid detection of nicosulfuron herbicide residues in field-grown maize is essential for implementing chemical remediation and optimizing spraying strategies. However, current detection methods are costly and time-consuming. This study analyzed residue levels in six maize varieties-both resistant and sensitive types-under two herbicide concentrations, categorizing residues into low, medium, and high levels. We developed the HerbiResNet model to predict and classify herbicide residues in maize leaves using spectral data. The model achieved a coefficient of determination (R²) of 0.88 for residue prediction and an accuracy of 0.87 for residue level classification on the test set, significantly outperforming traditional regression models (SVR, PLSR) and classical neural networks (MLP, AlexNet). Additionally, we explored combining spectral technology with deep learning, revealing strong correlations between specific spectral bands (around 550 nm, 680 nm, 750 nm, and 1000 nm) and herbicide residues as well as physiological changes in maize. This provides a solid theoretical foundation for the broader application of spectral technology in agriculture. Overall, the HerbiResNet model demonstrates substantial potential for precision agriculture and sustainable agricultural practices.