Estimation of chlorophyll content in cotton canopy leaves based on drone multispectral sensors: Multiple spectral information fusion

Accurate monitoring of chlorophyll content is of great significance for the sustainable development of precision agriculture. However, currently the widely used vegetation indices (VIs) for estimating field chlorophyll content are susceptible to external factors such as soil, vegetation, and weeds. Texture features can extract more information about crop growth from the spatial distribution of images to mitigate the limitations of VIs. Therefore, this study combines vegetation indices (VIs) extracted from remote sensing images with texture information (TFs) and utilizes four machine learning methods—partial least squares regression (PLSR), elastic net regression (ENR), random forest regression (RFR), and extreme gradient boosting regression (XGBoost)—to construct a model for inverting the chlorophyll content of cotton canopy leaves. The single model (VIs) was compared with the coupled model (VIs + TFs). The results indicate that the RFR model outperforms other models in terms of multi-indicator fusion feature processing (R² = 0.87). The RFR model demonstrated the highest accuracy with a low error rate, indicating that integrating spectral features is a key strategy for improving model accuracy. The results of 5-fold and 10-fold cross-validation, along with model evaluation metrics, showed that the random forest algorithm performed best when handling complex datasets, exhibiting strong stability and generalization capabilities. Therefore, it was selected as the optimal inversion model in this study. Explainable machine learning (SHAP) revealed the contribution of each indicator. This study demonstrates the effectiveness of integrating multispectral information in improving the accuracy of crop growth monitoring models. Based on RFR, a spatial distribution map of chlorophyll content was created. The research results provide a scientific basis for drone-based crop growth monitoring and field crop nutrient management. • The fusion of UAV multi-spectral information improves the accurate estimation of chlorophyll content. • Nonlinear regression model is the best choice for crop growth monitoring model. • 5-fold and 10-fold cross-validation increases feasibility. • The comprehensive model evaluation strategy provides a more powerful guarantee for ensuring the accuracy of the model.