Study On Inversion Method Of Soybean Growth Parameters Based On UAV Remote Sensing

Soybean,as one of the most important oil crops and food crops in China,plays an extremely important role in food security in China.Timely and accurate information of soybean growth can provide important basis for crop production management and early yield estimation.As an important parameter of crop population structure,leaf area index(LAI)can effectively reflect the changes of crop canopy structure,community life vitality and environmental effects,and is of great significance for crop growth evaluation and yield prediction.Aboveground biomass(AGB)is an important indicator of vegetation life activities,which can effectively reflect crop growth and is also an important functional index of ecosystem.In this paper,the soybean sown in early spring under the condition of dry farming in northern China was taken as the research object.The multi-spectral images of soybean pods and grain buds were obtained by carrying multi-spectral sensors on quadrotor UAV.Data were collected every 7-10 days,and LAI and AGB measured in the same period were combined.Ratio Vegetation Index(RVI),Difference Vegetation Index(DVI),Normalized Difference Vegetation Index(NDVI),Enhanced Vegetation Index(EVI),Triangular Vegetation Index(TVI),Soil-Adjusted Vegetation Index(SAVI),Optimization Soil-Adjusted Vegetation Index(OSAVI),Green Normalized Difference Vegetation Index(GNDVI),Atmospherically Resistant Vegetation Index(ARVI)and the best combination of VI obtained by stepwise regression were used to construct a single linear regression model,multiple linear regression model,BP neural network model and support vector machine(SVM)model for LAI and AGB in soybean pod setting stage and grain bulging stage.The main research results are as follows:(1)Unary linear regression method was used to construct the linear regression models of RVI,DVI,NDVI,EVI,TVI,SAVI,OSAVI,GNDVI,ARVI and LAI of soybean bulging stage,and to predict the sample data not involved in the modeling.Determinant coefficient(2R)and root mean square error(RMSE)were selected as the criteria for model evaluation.The results showed that the prediction effect of NDVI-LAI linear regression model was better,with2 R being 0.8149 and RMSE being 0.4045,followed by ARVI-LAI linear regression model and RVI-LAI linear regression model.(2)Multiple linear regression method was used to construct multiple linear regression models of soybean LAI and AGB,and the best VI combination screened by stepwise regression method was taken as the independent variables of the model,and LAI and AGB were taken as the dependent variables of the model.The results showed that: The multiple linear regression model of soybean bulging stage with NDVI,GNDVI and ARVI as independent variables had a better prediction effect(2R = 0.8704).The prediction effect of multiple linear regression model of AGB at soybean pod setting stage with RVI,OSAVI,GNDVI and ARVI as independent variables was better than that of multiple linear regression model of AGB at soybean bulging stage with NDVI,EVI and TVI as independent variables(2R = 0.7734 and RMSE =0.0777 kg/m~2).(3)BP neural network method was used to build the BP neural network model of soybean LAI and AGB based on the best VI combination obtained by stepwise regression,and the sample data not involved in the modeling were predicted.The results showed that: Taking the three neurons of NDVI,GNDVI and ARVI as input variables and LAI as output variables,the LAI prediction model2 R of soybean bulging stage was 0.8289 and RMSE was 0.3889.The prediction effect of the BP neural network model with 4neurons of RVI,OSAVI,GNDVI and ARVI as input variables and AGB at soybean pod setting stage as output variables is better than that of the BP neural network model with 3 neurons of NDVI,EVI and TVI as input variables and AGB at soybean bulging stage as output variables.Model2 R was 0.7722,RMSE was 0.0779 kg/m~2.(4)SVM method was used to construct and test the SVM model based on the optimal VI combination obtained by stepwise regression,with NDVI,GNDVI and ARVI as the input and LAI in the bulging period of soybean as the output.The verified model2 R was 0.8605,showing a good effect.The prediction effect of the SVM model with RVI,OSAVI,GNDVI and ARVI as inputs and AGB at soybean pod setting stage as outputs was better than that of the SVM model with NDVI,EVI and TVI as inputs and AGB at soybean bulging stage as outputs.The tested model2 R was 0.786 and RMSE was 0.0755 kg/m~2.(5)The results showed that the multiple linear regression model with NDVI,GNDVI and ARVI as independent variables had the best effect on the prediction of soybean LAI at the bulging stage,with Model2 R being 0.8704.SVM model with RVI,OSAVI,GNDVI and ARVI as input values had better prediction effect on AGB of soybean pods than other models,with2 R being 0.786 and RMSE being 0.0755 kg/m~2.The SVM model with NDVI,EVI and TVI as input values had the best prediction effect on AGB of soybean bulging stage,with2 R being 0.7637,RMSE being 0.0957 kg/m~2.Multiple linear regression models were used to invert LAI from spectral images at the bulging stage of soybeans,and SVM model was used to invert AGB from spectral images at the pod setting stage of soybeans.