Feature selection of wheat field biomass model retrieved by UAV remote sensing
The unmanned aerial vehicle(UAV)multispectral technique is a popular method for rapid and nondestructive de-termination for field biomass(AGB)of wheat.However,the multispectral method usually produces a large number of highly correlated repetitive features in the calculation of vegetation features,so it is of great significance to features selection and determine the model with simple structure and high precision.In this paper,a hybrid coded Grey Wolf particle swarm opti-mization(CGWOPSO)algorithm was proposed,which can achieve both feature screening and parameter optimization.To e-valuate the performance of this method,the performance of two popular feature selection methods(Pearson and SHAP meth-ods)driven by Extreme gradient boosting model(XGBoost)for AGB was compared.The results show that the AGB model based on the SHAP method yield RMSE 3.0％to 16.3％lower than the Pearson method.The accuracy of CGWOPSO-XGB model was higher than that of XGB model based on SHAP method,and its RMSE is 16.0％lower than that of the latter.
hybrid codingGrey Wolf particle swarm optimization algorithmSHAPfeature selectionvegetation index
国家科技期刊平台
NSTL
万方数据
