Inappropriate use of herbicides in agricultural fields can result in significant phytotoxic effects on maize, adversely affecting both yield and quality. The development of herbicide-resistant maize varieties is a crucial strategy for managing herbicide-related challenges in agriculture. However, traditional methods for screening resistant strains are often time-consuming and inefficient, presenting several limitations. Consequently, there is a pressing need for a rapid and precise methodology to evaluate herbicide resistance in maize. This study involved a field trial to evaluate herbicide resistance in maize, utilizing two resistant and two sensitive varieties. We developed an ATT-RES model using hyperspectral data from maize leaves subjected to herbicide stress to predict resistance. The model’s accuracy and generalizability were confirmed by integrating datasets across different years, planting seasons, and varieties. Additionally, this study examined the correlation between leaf spectra and corresponding physiological and biochemical indicators under herbicide stress by constructing a regression model and investigated the mechanisms through which spectra can predict herbicide resistance in maize. The findings indicate that on the fourth day after herbicide treatment, the model’s predictive accuracy exceeded 90 %. The accuracy peaked on the seventh day, achieving up to 95 %. The ATT-RES-based transfer model attained prediction accuracy of 93.55 %, 90.63 %, and 95.86 % across various years, varieties, and seasons, respectively. The results presented here offer significant advantages over traditional models such as Support Vector Machines (SVM), Multilayer Perceptrons (MLP), and AlexNet. Furthermore, we employed Partial Least Squares Regression (PLSR) to analyze the relationship between leaf physiological-biochemical parameters and spectral data. The analysis revealed significant correlations between the spectra and key biochemical markers: glutathione (GSH, Rp2 = 0.59), malondialdehyde (MDA, Rp2 = 0.48), chlorophyll content (SPAD, Rp2 = 0.82), and water content (WC, Rp2 = 0.71). The prediction of resistance was primarily influenced by variations in leaf SPAD and WC, but also showed correlations with GSH and MDA levels in the leaves. This study contributes valuable theoretical and technological insights that facilitate early and precise prediction of herbicide resistance in maize. The advancements demonstrated in this research hold significant potential for improving crop management and reducing the adverse impacts of herbicides on the environment.
Read full abstract