基于多光谱与热红外影像的烤烟地土壤含水率反演OA

In order to achieve rapid and precise monitoring of soil moisture content in flue-cured tobacco fields,this study was conducted in nine flue-cured tobacco experimental plots in Tuokeng Village,Menggu Town,Qiaojia County,Zhaotong City,Yunnan Province as the study area.The research integrated multispectral and thermal infrared UAV imagery with machine learning models to retrieve soil moisture content.Data acquisition involved multispectral imagery from a DJI Mavic 3 UAV(four bands,spatial resolution of 0.003 m)and thermal infrared imagery from a DJI Matrice 4T UAV(VOx sensor,spatial resolution of 0.005 m),along with in-situ measurements of soil volumetric water content.The images were preprocessed using ENVI and Pix4D software,followed by the extraction of nine key multispectral features.These were combined with thermal infrared DN values to construct the feature set.A series of machine learning models—including Random Forest(RF),Support Vector Machine(SVM),Extreme Gradient Boosting(XGBoost),and Convolutional Neural Network(CNN)—were applied to develop both single-source and multi-source soil moisture retrieval models.Model performance was evaluated based on root mean square error(RMSE)and the coefficient of determination(R2).The results demonstrated that multi-source models significantly outperformed their single-source counterparts.The RF model,incorporating multispectral features and thermal infrared DN values,yielded the best performance(RMSE=1.30%,R2=0.79),with XGBoost providing comparable results(RMSE=1.31%,R2=0.78).SHAP analysis further revealed that the Normalized Difference Red Edge Index(NREI),NDVI_RVI,and thermal infrared DN values were the key features.This study highlights the potential of combining multispectral and thermal UAV imagery with machine learning for efficient soil moisture retrieval in flue-cured tobacco fields,providing technical support for precision irrigation and sustainable water resource management in agricultural regions.