177 / 2024-03-11 12:26:45

Inversion of Soil Water Content in UAV Multispectral Remote Sensing Based on Feature Variable Screening

soil moisture content; UAV; XGBoost feature screening; optimal subset screening; partial least squares regression; random forest regression.

10、绿色低碳转型与可持续发展 > 4、科技创新发展与碳中和实现

Soil water content is one of the important factors affecting the growth of crops, and plays an important role in crop yield estimation and drought monitoring. In soil water content calculation, multiple spectral variables are generally extracted for inversion, but the spectral information contained between the variables may have redundancy and overlap, in order to extract effective feature variables and make them independent of each other, the thesis researches on the feature variable screening method and verifies the applicability in soil water content inversion. Based on the UAV multispectral images, 12 types of vegetation indices such as Normalized Difference Vegetation Index (NDVI) are calculated, combined with the UAV thermal infrared (TIR) data to calculate the surface temperature LST and the corresponding TVDI, as well as four backscattering coefficients obtained from miniSAR data processing, and the XGBoost feature variables and the optimal feature variables are selected by the method. XGBoost feature variables and the Best Subset Selection (BSS) algorithm were used to screen the optimal variable combinations, and then Partial Least Squares Regression (PLSR) and Random Forest Then, the soil water content at the tasseling stage of winter wheat in the experimental area was inverted using Partial Least Squares Regression (PLSR) and Random Forest Regression (RFR) algorithms.  The inversion accuracy of XGBoost-PLSR model is better than that of XGBoost-RFR at 0-20 cm depth, while the opposite is true at 0-20 cm depth, and the inversion accuracy of BSS-RFR model is higher than that of BSS-PLSR at 0-20 cm depth.The results of the research can provide theoretical and technical support for the inversion of soil water content by UAV multispectral remote sensing, and provide a test basis for the monitoring of soil moisture on a large scale by satellite remote sensing. The research results can provide theoretical and technical support for UAV multispectral remote sensing inversion of soil water content, and provide test basis for satellite remote sensing large-scale soil moisture monitoring.