Estimation Methods Of Maize Yield Based On Feature Weighting

Maize is the largest food crop in China.Timely and accurate monitoring of maize growth and estimation of maize yield information can provide decision support for maize planting management,which is of great significance to food production.Compared with satellite remote sensing platform,UAV remote sensing platform has the advantages of easy operation,high spatial resolution and low cost.As an important indicator of crop growth,leaf area index(LAI)has a strong correlation with crop yield.At present,the common LAI remote sensing estimation methods for crops ignore the temporal variation characteristics of crop LAI.At the same time,the common crop yield remote sensing estimation methods have less discussion on the impact of various parameter characteristics of each growth period on yield.In view of the above two problems,this study starts from two parts:the LAI estimation method based on temporal feature weighting based on Shapley strategy and the yield estimation method based on multi growth periods feature weighting based on decision tree.Finally,a high-precision LAI estimation and yield estimation method of maize is obtained.The main research contents and results are as follows:(1)LAI estimation method based on time series feature weighting.Different growth mechanisms in different growth stages of crops lead to different growth changes in different stages.Similarly,crop LAI changes also have time-series characteristics,but it is difficult to quantify the time-series characteristics of LAI into specific parameters into the model.Based on the Shapley strategy of cooperative game theory,a combined LAI estimation model is constructed.The combined model quantifies the contribution of each growth period sub models through the weight coefficient,and introduces the temporal characteristics of LAI in each growth period.The experimental results of Lai estimation show that compared with the LAI estimation model directly based on the characteristic parameters of the whole growth period,the combined LAI estimation model realized by Shapley strategy can integrate the temporal variation characteristics of maize LAI and optimize the estimation accuracy of LAI.Among them,XGBoost-Shapley model(R~2:0.97,RMSE:0.021,RPD:6.9)has the best performance in LAI estimation.(2)Yield estimation method based on multi growth period parameters weighting.In order to explore the influence of different parameter characteristics of each growth period on yield estimation,firstly,the corn LAI data in the study area are obtained based on the XGBoost-Shapley combined LAI estimation model,the maize canopy temperature is extracted based on visible light image and thermal infrared image by using the improved Canny edge detection algorithm,and then the influence of each growth period on yield is analyzed through the yield estimation experiment of single growth period,XGBoost and RF were used to quantify the contribution of each characteristic parameter to the final yield in each growth period,and a multi growth period parameter weighted yield estimation model was constructed.The experimental results of yield estimation show that the XGBoost-Weighting(R~2:0.9792,RMSE:102.337kg*hm^-2,RPD:7.43)yield estimation model based on multi growth period parameters weighting method can more accurately reflect the accumulation process of maize organic matter than the single growth period yield estimation model.The parameter weighting scheme is in line with the mechanism characteristics of maize growth process,and the yield estimation effect of the model has been significantly improved.To sum up,an accurate and efficient LAI estimation and yield estimation scheme of maize was obtained.It can realize high-precision LAI estimation and yield estimation of medium and large-scale maize fields based on UAV remote sensing platform,and provide feasible technical reference for maize growth monitoring and yield estimation.