Research On Maize Aerial Biomass Estimation Method Based On Canopy Multispectral Image

Maize(Maize)is the largest crop planted in our country,followed by wheat and rice.In recent years,with the increase of corn planting area and yield,the growth of corn has been paid more and more attention,and the prediction of aboveground biomass related to growth has also become the focus.Aboveground biomass(AGB)is a parameter used to reflect crop phenotype and effectively monitor crop growth.It is an important indicator to describe vegetation growth state and has great significance for predicting crop growth trend and yield.The traditional AGB measurement method requires manual measurement and is difficult to collect.In order to quickly and accurately know the AGB and growth trend of maize,it is necessary to predict the form of AGB based on canopy multispectral images.This is our current focus of corn research.Based on field maize experiments in Ainishan Township,Shuangbai County,Chuxiong Prefecture,Yunnan Province,and Mengjie Town,Mengjie Town,Manshi City,Dehong Prefecture,Yunnan Province during 2022-2023,this study used DJI Sprin-4multispectral UAV(P4M)to obtain remote sensing images of maize at the joint,silking and grouting stages in the experimental areas.The measured aboveground biomass was obtained by randomly obtaining five maize plants in each plot,and five vegetation indices were selected(ratio vegetation index RVI,difference environmental vegetation index DVI,normalized difference vegetation index NDVI,enhanced vegetation index EVI,and optimized soil-regulated vegetation index OSAVI).Partial least square method,ridge regression algorithm and BP neural network were used to invert AGB of maize in the test area.The research contents were as follows:(1)During the execution of tasks,the pictures taken by UAV will be affected by various factors,resulting in the problem of image distortion.In order to solve this problem,the parameters of UAV flight path are set.By setting flight altitudes of 15 meters,25meters and 35 meters in the field,the optimal flight parameters were determined as follows:flight height of 15 meters,course overlap of 80%and side overlap of 70%,the remote sensing image with higher definition could be obtained.(2)Based on the multi-spectral images,partial least squares algorithm was used to predict the biomass,and the vegetation index reflectance and aboveground biomass construction models were constructed under the observation of jointing stage,spinneret stage and grout stage.Among the prediction results,partial least square algorithm was used to construct the measured value-DVI predicted value model at grout stage with the highest accuracy,the coefficient of determination R~2was 0.917,the root mean square error RMSE was 16.99g/m~2,and the normalized root mean square error NRMSE was 28.31%.The experimental results showed that the partial least squares model could accurately predict the biomass of the upper part of the corn field in the filling stage.(3)The ridge regression algorithm was used to predict the biomass based on multi-spectral images.The ridge regression model was constructed by exponential fusion of EVI,OSAVI,DVI,NDVI and RVI as the input of the model and the aboveground biomass as the output.Among the prediction results,Ridge regression algorithm was used to construct the prediction result model of spinneret stage with the highest accuracy,the coefficient of determination R~2was 0.89,the root mean square error RMSE was 12.03g/m~2,and the normalized root mean square error NRMSE was 14.73%.The experimental results showed that the ridge regression model could accurately predict the biomass of the upper part of corn field in the silking stage.(4)Based on multi-spectral images,BP neural network algorithm was used to achieve biological prediction,and a model of exponential fusion of five planting covers and aboveground biomass was constructed.Among the prediction results,the prediction model of grouting stage has the highest accuracy,the determination coefficient R~2is 0.928,the root-mean-square error RMSE is 21.32g/m~2,and the normalized root-mean-square error NRMSE is 9.84%.The accuracy and recall rate of aboveground biomass training set were90.31%and 86.47%respectively.The experimental results showed that BP neural network could accurately predict the biomass of the upper part of corn field in the filling stage.(5)Through comparative analysis of model results,the accuracy of modeling results and prediction results reached 89.5%and 84.9%in jointing stage.The accuracy of the modeling results and the prediction results reached 86.6%and 91.4%respectively.The accuracy of modeling results in grouting stage reached 85.5%,and the accuracy of forecasting results in grouting stage reached 92.6%,which was higher than other models.The BP neural network was determined to be the best inversion model,and AGB inversion was carried out on the remote sensing images of maize in three periods to obtain the spatial distribution map of aboveground biomass.The results show that the UAV multi-spectral image can well estimate the biomass of the upper part of the corn field,and using the estimated aboveground biomass inversion spatial distribution map,predicting aboveground biomass can help farmers estimate their yield,and effectively promote the research and application of remote sensing theory of crops in China.