The Research Of Maize Yield Estimation Affected By Lodging Based On Remote Sensing Data

Maize is the largest food crop in the world and also in China,which plays an important role in ensuring national food security.Accurate and nondestructive estimation of maize yield in large area and low cost is very important for food policy making and food security.There is a strong correlation between maize yield estimation and the characteristic indexes extracted from remote sensing images.Different yield estimation models directly affect the accuracy and precision of maize yield estimation.Current yield estimation methods are mainly based on field investigation and farmers’ professional knowledge,crop growth models and remote sensingbased methods,and integrated crop growth models combining environmental factors(hydrology,meteorology,soil,etc.)and remote sensing factors.The development of new technology improves the precision and timeliness of yield estimation by remote sensing,which makes the yield estimation model based on the traditional single specific growth period to the refined monitoring based on the whole growth period as far as possible,which is of great significance to the crop growth monitoring and the study of crop yield formation.In the actual production process,maize lodging is one of the most destructive natural disasters during maize growth period,which not only affects maize grain quality,but also causes serious yield reduction.After lodging,the spectrum and canopy structure of maize would be seriously changed.The traditional single-temporal remote sensing estimation method has a low accuracy and cannot meet the needs of production process.Therefore,it is important to estimate maize yield by combining the time-series results of multi-temporal and multi-characteristic indexes in the case of lodging to improve the estimation accuracy and calculate lodging loss.In this paper,the maize experimental field located in Xinxiang experimental base of Chinese Academy of Agricultural Sciences in Henan Province was selected as the research area,and the experimental areas with different sowing dates,varieties,densities and nitrogen application rates were selected as the research objects.Based on the UAV remote sensing images and the physiological indexes measured on the ground,the image processing,quantitative remote sensing and statistical analysis methods were used to analyze the correlation between the maize growth in order to realize the classification of lodging plots,the identification of key phenological periods and the construction of time series canopy volume yield estimation model,the influence of lodging characteristic factors on yield estimation was analyzed.The main contents of this study are as follows:(1)Evaluation and influencing factors of maize lodging based on UAV RGB imageIn this study,a set of high-precision classification methods for lodging grade of maize were established by using spectral,texture,canopy structure and other characteristic indexes,respectively,using pixel-based supervised classification and object-oriented classification methods.The results show that the best supervised classification method based on pixels is random forest.Using object-oriented classification,using random forest classification method combined with the original image,the canopy surface model and texture characteristics of the highest overall classification accuracy,the characteristics of the index selection is the best solution to the original RGB images,the change of the canopy surface model and vegetation index or texture features are combined,the overall accuracy is 86.96%,the Kappa coefficient is 0.7931.Combined with the spatial distribution of lodging degree and different experiments,the influencing factors of lodging degree were preliminarily analyzed.Sowing date experiments showed that R1(silking stage),VT(complete tasseling stage)and V14(14 leaf stage)were three stages for lodging.The results of nitrogen experiment showed that lodging resistance of maize could be improved with the increase of nitrogen application or at the critical growth stage.Varieties experiments showed that Liaodan 585,LP68 and Zhongdan 909 had the best lodging resistance in this study.In addition,the density test showed that the lodging resistance of different maize varieties depended on the planting density,and the higher the planting density,the greater the lodging risk.In conclusion,these results can quickly determine the lodging degree of maize in the field,thus improving the field management strategy.(2)Monitoring of key phenological phases of MaizeWith the help of unmanned aerial vehicle remote sensing platform,this study obtained crop canopy structure,spectrum,texture,and photosynthetic and nutrient indexes related to crop growth.Combined with images near the ground,the true description of maize growth was realized.Based on this,the optimal vegetation index--GRVI was selected to construct the growth curve of maize.Then,after Savitzky-Golay filtering on growth curve interpolation data for day by day,and from the curve of the filter after extraction combined with dynamic threshold method in field maize seedling stage,tasseling stage,physiological ripeness and length of growing season(emergence to physiological mature)and so on four important phenophase,one of the highest accuracy is tasseling stage,the error is within two days,the seedling stage and physiological mature period is low,on the basis of the experimental plot four phenological phase space mapping.(3)Maize yield estimation based on time series weighted canopy volume modelOn the basis of phenological phase extraction,different types of time series weighted canopy volume models were constructed by combining lodging grade and green canopy coverage,plant height,vegetation index,texture characteristics and other index curves related to maize,and then yield estimation models based on single or multi temporal cumulative values were established for experimental maize plots.Using partial least squares regression,support vector regression,random forest regression and other machine learning methods to estimate yield,this method can screen out the best yield estimation period is the cumulative value of the characteristic curve of vegetative growth stage,the best yield estimation model is the texture weighted canopy volume model,the best estimation method is the random forest method,and the lodging characteristic index can be used in the yield estimation The optimal modeling accuracy can reach 82%,and the RRMSE is 11.21%.