| Using scientific fertilization methods,exploring a variety of field observation methods and combining agronomy knowledge to strengthen the accurate perception of crop nutrient information,monitoring crop nitrogen nutrition status in real time,and formulating scientific nitrogen management strategies according to the nutrient demand laws of crop growth are of great significance for improving crop nitrogen utilization efficiency,increasing farmers’economic benefits,and improving environmental conditions.The sensor of UAV remote sensing platform is limited by its load capacity,and it needs to meet certain requirements in precision,quality and size.Different sensors mounted on it for farmland monitoring are less affected by weather,and can obtain high-resolution remote sensing images,which can meet the requirements of auxiliary farmland precise management.In this study,the experimental field of Sankeshu Village in Siping City is taken as the research area,spring corn is taken as the research object,Mini-MCA multispectral camera is used to acquire multispectral remote sensing images of canopy of spring corn in the experimental area with different treatments in three years.Traditional empirical regression,stepwise multiple regression and three traditional machine learning algorithms(support vector regression,BP neural network and random forest algorithm)are used to retrieve nitrogen nutrition parameters of spring corn such as aboveground biomass,nitrogen concentration,nitrogen uptake and nitrogen nutrition index.The main work and conclusions are as follows:(1)In the process of UAV multispectral remote sensing image processing and remote sensing variable extraction,based on the method of combining empirical linear correction and illumination sensor correction,the calibration file of the sensor is corrected again by using the program written by Python 3.2 software,taking the canopy reflectance obtained by ASD ground high spectrometer as a reference;The corrected multispectral images were preprocessed by band merging,image splicing and image clipping,and finally the corn canopy multispectral orthographic images were obtained.(2)Nitrogen application rate and planting density have obvious influence on various agronomic parameters of spring maize.Adding growth period,nitrogen application rate and planting density as variable factors into the construction of machine learning estimation model can increase the estimation ability and verification accuracy of the model,facilitate the adjustment and optimization of the estimation ability of the model,and improve the applicability of the inversion model.(3)The best multispectral modeling method for unmanned aerial vehicles is random forest algorithm.The diagnosis model of maize nitrogen nutrition parameters constructed by traditional machine learning algorithm is better than traditional empirical regression model and multiple stepwise linear regression model.However,the order of the precision of the inversion model based on three machine learning algorithms is random forest>BP neural network>support vector regression.the maximum R~2values of the precision prediction set of the inversion model based on random forest algorithm are 0.93,0.91,0.85 and 0.83 respectively,which are3.39%,1.87%,0.79%and 2.70%higher than the maximum precision of the BP neural network model respectively.Compared with the maximum precision of SVR regression model,the precision of random forest algorithm is increased by 9.45%,24.54%,10.62%and 4.77%respectively.Compared with traditional empirical regression and multiple stepwise linear regression,random forest algorithm is more obvious. |
