Research On Monitoring Of Dry Farmland Millet Phenotypic Information Based On Unmanned Aerial Vehicle Remote Sensing

The monitoring of crop phenotype information based on UAV remote sensing technology has the characteristics of rapid and accurate.Research on this aspect of millet is less in modern agriculture and is an important aspect of crop phenotype monitoring.This paper selects farmland in Wujiabao Village,Taigu District,Jinzhong City,Shanxi Province as the research area.UAVs are equipped with high-definition digital cameras and multispectral cameras.The Wujiabao experiment will be collected from June to September2020 at a height of 30 m from the ground.Visible light images and multi-spectral images of millet planted in the field,to monitor plant height,leaf area and biomass information,in order to provide some reference value for the development of field growth monitoring technology for millet planting in the future.The main results of this paper are as follows:(1)Use drones to obtain high-definition digital images,generate digital elevation model DEM,and analyze the correlation between millet artificial measurement of plant height and sensor-based measurement of height.Correlation analysis between the normalized value of the elevation data of the digital elevation model and the normalized value of the real plant height showed that the correlation between July 15 and August19 was extremely significant,and the coefficient of determination R~2 was 0.698 and 0.87,respectively.The four varieties in the study area were distinguished and analyzed separately.Among them,the correlation of the four varieties was extremely significant.The coefficient of determination R~2 after the analysis of mutant A was 0.886;the coefficient of determination R~2 after the analysis of mutant B was 0.972;the coefficient of determination R~2 after the analysis of body C was 0.954;the coefficient of determination R~2 after the analysis of Jin Gu 21 was 0.965.This study proved the possibility of UAV remote sensing to monitor millet plant height.(2)Use the multi-spectral camera carried by the drone to obtain the image to generate the digital orthophoto DOM,and use the tool to extract the spectral information of the quadrat in the image and calculate the 13 planting index,and the spectral information as the feature parameter to select the optimal feature Combine,and then use three algorithms including random forest,support vector machine,and linear regression to construct the prediction model of cotyledon leaf area and biomass.For biomass data,in the case of distinguishing varieties,a composite model was established,and it was decided to use random forest and linear regression for mutant C and Jin Gu 21 to separately establish biomass prediction models.However,the model established by random forest is used as the biomass prediction model for predicting mutant A and mutant B without distinguishing between varieties.The coefficient of determination R~2 for the fit between the predicted value of the composite model and the true value is 0.86,the root mean square error RMSE is 0.124,and the effect is the best.For leaf area data,in the case of distinguishing varieties,a composite model is established,and the leaf area prediction model of mutant A is constructed based on the support vector machine algorithm;the leaf area prediction model of mutant B is constructed using linear regression;the mutation is constructed using random forest algorithm the leaf area prediction model of body C;the leaf area prediction model of Jin Gu21 is constructed using support vector machine algorithm.The coefficient of determination R~2 between the predicted value of the composite model and the true value fitting is 0.82,and the root mean square error RMSE is 0.422,which has the best effect.