Monitoring Of Nitrogen Accumulation In Winter Wheat Plants Based On UAV Multispectral And GF-2 Remote Sensing Images

Rational management of nitrogen fertilizer is a prerequisite for modern agriculture.Only with sufficient nitrogen supply can high-quality wheat and optimal yield be obtained.In order to realize the rapid and non-destructive monitoring of winter wheat nitrogen nutrition,the multi-spectral images of 30 m winter wheat canopy during the main growth period were obtained by UAV equipped with a multi-spectral camera,and the effects of multi-source information extracted by UAV on the estimation model of winter wheat plant nitrogen accumulation were explored.Secondly,the effects of uav flight height on nitrogen accumulation estimation model of winter wheat were further discussed by obtaining the vegetation index and texture characteristics of uav images at different altitudes(flight altitude of 30,60 and 90 m,resolution between 1.05-3.26cm).Finally,in order to achieve region-wide monitoring of plant nitrogen accumulation,the estimation results of winter wheat plant nitrogen accumulation were compared between 30 m UAV multispectral data and GF-2 satellite remote sensing data,and the applicability of the estimation model of plant nitrogen accumulation constructed by UAV multispectral data in GF-2 image was discussed.The above research provides technical support for monitoring nitrogen accumulation of winter wheat plants at field and regional scales.The main results of this study are as follows:(1)Based on winter wheat jointing stage,flowering period and grouting period 30 m UAV multispectral image,extraction of digital orthogonal projection as spectral information(including canopy reflectance and vegetation index)and texture characteristics and combined with digital surface model extracted plant height information,using multiple linear regression,partial least-squares regression and BP neural network method to build winter wheat plant nitrogen accumulation estimate model,To explore the effects of different data sources on nitrogen accumulation estimation of winter wheat.Spectral information,texture characteristics and extracted plant height were correlated with nitrogen accumulation of winter wheat plants to varying degrees.When digital ortho image information was used to estimate the nitrogen accumulation of winter wheat plants,the R~2,RMSE and RPD models constructed by spectral information and texture features ranged from 0.712-0.824,0.305-2.233 g/m~2 and 1.860-3.331,respectively.The BP neural network model at flowering stage had the best effect(R~2 = 0.824,RMSE = 0.305 g/m~2,RPD = 3.331).The estimation accuracy of nitrogen accumulation of winter wheat was improved when combined with the extracted plant height.Therefore,texture feature +extraction plant height increased most obviously.The BP neural network model constructed by spectral information + texture feature + extraction plant height at flowering stage had the best estimation effect(model R~2 = 0.828,RMSE = 0.154 g /m~2,RPD = 3.399).Make full use of digital orthogonal projection image and digital surface model two kinds of data source information,combined with machine learning algorithms,build the growth period of winter wheat plant nitrogen accumulation estimate model,can effectively improve the effect of winter wheat plant nitrogen accumulation quantity estimation,do a specific location for farmers to adjust crop into management decision-making,guarantee food security is of great significance.(2)UAV remote sensing images with different resolutions at jointing stage,flowering stage and filling stage were obtained by setting flight altitude of 30,60 and 90 m to explore the effect of UAV flight altitude on prediction model of nitrogen accumulation of winter wheat plants.Vegetation index based on screening and texture characteristics,the use of partial least-square regression and BP neural network to set up the vegetation index,texture characteristics and vegetation index + texture characteristics of winter wheat plant nitrogen accumulation prediction model,and the model at different heights for cross validation,the R~2,RMSE and RPD index for the stability of the model are analyzed.The results show that the two methods have the best stability of the prediction model based on vegetation index and texture feature extracted from remote sensing images at a height of 30 m.The R~2,RMSE and RPD of the three models were0.517-0.842,1.274-4.157 g/m~2 and 1.372-3.050,respectively.The stability of the model constructed by BP neural network under the three kinds of modeling information was better than that of partial least squares regression.The R~2 and RPD of the verified model increased from0.009-0.388 and 0.046-1.438 respectively,and the RMSE decreased from 0.079-8.527 g/m~2.The order of stability of the prediction model of n accumulation was vegetation index + texture feature > vegetation index > texture feature.The prediction accuracy of plant nitrogen accumulation can be improved by integrating vegetation index,texture feature and vegetation index + texture feature from multi-height remote sensing images(the range of model R~2,RMSE and RPD are 0.867 ~ 0.937,1.320 ~ 1.611 g/m~2 and 3.368-3.810 respectively).Therefore,under the condition of both efficiency and accuracy,the appropriate increase of uav flight height and comprehensive utilization of vegetation index and texture features can achieve a better prediction effect on plant nitrogen accumulation.(3)Due to its sub-meter resolution,GF-2 satellite images provide an important technical means for monitoring the growth of winter wheat.In order to achieve accurate monitoring of nitrogen accumulation in winter wheat plants in a large range,it is necessary to build a monitoring model based on abundant ground data.In this study,the canopy images of winter wheat under different nitrogen application levels were obtained by uav platform,and the vegetation index of UAV multispectral images and GF-2 satellite images were compared to explore the accuracy and stability of the estimation model of winter wheat plant nitrogen accumulation based on UAV multispectral vegetation index in GF-2 inversion.The results show that the reflectance of UAV in all spectral bands has a high correlation with the reflectance of GF-2 in Blue,Red and Nir bands,and the R~2 order is Nir > Red > Blue.The correlation coefficients between the vegetation indices of the two sensors and the nitrogen accumulation of winter wheat ranged from 0.515-0.614(GF-2)and 0.644-0.862(UAV),respectively,in which the correlation coefficients of RDVI had the greatest difference and that of OSAVI had the smallest difference.By comparing the prediction results of the two vegetation index construction models,it can be seen that the UAV multispectral vegetation index OSAVI has the best image applicability at two scales,and the correlation of UAV scale and satellite scale inversion results reached 0.695 respectively,and reached extremely significant correlation.Therefore,it is feasible to estimate plant nitrogen accumulation in a large range by simulating satellite remote sensing bands with uav remote sensing information in a small range.