Research On Maize Coverage Estimation And Yield Inversion Method Based On Ensemble Learning

Corn is one of the most important food crops in China.How to estimate corn coverage and yield timely and accurately has become a research hotspot in recent years.UAV has the characteristics of high flexibility and high spatial resolution.It is of great significance to use UAV to monitor crop growth and obtain crop growth information.This study is based on the corn experimental area in Zhaojun Town,Dalate Banner,Ordos City,Inner Mongolia Autonomous Region.While carrying out the corresponding ground data collection work,it uses UAV to collect high-definition multispectral remote sensing images to preliminarily explore the estimation of corn coverage,and then uses multi temporal information to establish a corn yield estimation model.Finally,a more complete corn coverage estimation and yield inversion method was obtained.The main research contents and results are as follows:(1)Research on estimation method of corn coverage.Based on UAV multispectral images,9 vegetation indices were extracted as feature variables,and the redundant features were removed by Pearson correlation coefficient and random forest reverse validation weight factor,and the best feature combination was selected to estimate corn coverage.Random Forest,Gradient Lifting Tree,Support Vector Regression and Ridge Regression are used as primary learners,and ridge regression is used as secondary learners.Adaptive enhancement algorithm is used to enhance the effect of base model,and corn coverage estimation model based on Stacking ensemble learning is established.Cross validation is used to further improve the generalization ability of the model.The model takes the measured data of the current year as reference Line accuracy evaluation,robustness evaluation based on the next year’s data.The results showed that the R~2 of Stacking ensemble learning model was 0.9588,RMSE,MAE and MAPE were 0.0378,0.0307 and 4.26%respectively.Compared with other ensemble learning models,R~2 of Stacking ensemble learning model increased by 0.0496,RMSE,MAE and MAPE decreased by 0.0239,0.0149 and 2.90%respectively.MAPE was 5.82%.(2)Research on maize yield inversion method based on coverage.Based on the multi temporal remote sensing information and corn coverage monitoring data,Random Forest,Support Vector Regression and Simple Linear Regression were used to inversion the regression model of corn yield estimation.At the same time,it is found that the performance of corn coverage monitoring data combined with weighted coverage is the best.Finally,the model is selected for corn yield inversion and mapping.The results show that the R~2 of Random Forest yield inversion model is 0.9029,RMSE is 271.2787,MAE is 335.2675,and MAPE is only 3.65%.Compared with Support Vector Regression method,MAPE is reduced by 0.94%,and compared with Simple Linear Regression method,MAPE is reduced by 3.99%.In this research,a more accurate method of maize coverage estimation and yield inversion was obtained,which can provide a new idea for rapid diagnosis and evaluation of crop growth and yield estimation,and provide effective support for the realization of smart agriculture and precision agriculture.