Answers are grounded in this paper and cite their sources.
Cotton spider mite damage threatens stable cotton production, and rapid, non-destructive field monitoring is therefore important for timely control. Although unmanned aerial vehicle (UAV) remote sensing provides an effective approach for plot-scale pest detection, methods based on single-source data often fail to capture the full range of damage-related information and may show limited robustness. In this study, a quadrotor UAV equipped with a 10-band multispectral sensor and an RGB camera was used to collect field imagery during the outbreak period of cotton spider mite infestation. Vegetation indices, canopy texture features, and hue, saturation, and value (HSV) color features were extracted and combined with quadrat-based field survey data to construct plot-level damage index (DI) labels and a multisource feature dataset. Pearson correlation analysis and category-wise feature screening identified six core features, namely DVI, SAVI, GI, R_Entropy, R_Second Moment, and Hue. These features were then used to evaluate four machine learning regression models: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Support Vector Regression (SVR). Among them, RF performed best on the test set, with R 2 = 0.8859 , RMSE = 0.1641, and MAE = 0.1365. Compared with the best model based on single-source vegetation-index features, the fusion model showed a higher R 2 and lower RMSE and MAE, with improvements of 0.0562, 0.0366, and 0.0116, respectively. These results indicate that multisource feature fusion improves cotton spider mite damage index estimation and provides useful support for precision pesticide application in cotton fields.
No takes yet. Share an insight, caveat, or question.
Han et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: