Typical Crop Recognition And Water And Fertilizer Monitoring Based On UAV Multi-spectral Remote Sensing

Crop plant water content is a key parameter for the diagnosis of crop water deficit and the formulation of irrigation water demand.chlorophyll content is an important indicator to reflect the physiological state and growth trend of crops.Crop identification is a prerequisite to accurately grasp the basic information of agricultural situation and is also the basis of agricultural remote sensing application.UAV multi-spectral remote sensing has the advantages of high efficiency and flexibility in crop identification and water and fertilizer monitoring.It is an emerging research method applied in the field of agriculture.Winter wheat as the research object,based on this,this paper takes the low-altitude uav platform with multi-spectral camera UAV image to crops,observation of crop canopy spectral characteristics of ground investigation crop types at the same time,collection of winter wheat under different water treatment plant moisture content and SPAD value data,selects the threshold segmentation based on spectral variables of decision tree classification of plant extract,application of one dollar and multiple regression analysis method based on spectral reflectance or spectral vegetation index of winter wheat plant moisture content and SPAD value estimation model.The main results are as follows:(1)The phase and spectral characteristics of the image were analyzed by determining the types of features of interest,and then the normalized vegetation index NDVI,normalized green-blue difference index NGBDI,modified ratio vegetation index MSR,and red edges were determined.The band reflectance can be used as the optimal classification feature.The decision tree classification method based on the threshold segmentation of spectral variables is used to implement the feature classification,extract the planting area,and select the ground survey data based on visual interpretation for method verification.The results show that the method of decision tree classification based on temporal and spectral features is effective,and it is used to extract wheat,fruit trees and greenhouses.From the classification effect map and confusion matrix of decision tree,the overall accuracy of classification is 88%,and kapaa coefficient is 0.83.The results show that the overall classification effect of the method is good.Based on the above experimental results,it is considered that the method is based on temporal and light The decision tree classification method of spectral features can be applied to the extraction of ground feature information from multi spectral image of UAV.(2)Constructed a spectral reflectance model and a spectral vegetation index model,and screened the interpretation models of winter wheat plant moisture content in typical regions.The moisture content of winter wheat plants was significantly correlated with the reflection spectrum at the level of 0.05,and the band sensitive to the moisture content of the plants was the near-infrared band with a center wavelength of 840 nm,which showed a very significant negative correlation.The spectral indexes more sensitive to the moisture content of winter wheat plants at the jointing stage are SAVI and EVI.The spectral indexes more sensitive to the moisture content of winter wheat plants at the heading stage are NDVI,SAVI,EVI,and SR.The spectral indices more sensitive to the moisture content of winter wheat plant at the grain filling stage are NDVI and SR.Among the models established based on multiple regression analysis,the spectral vegetation index model is superior to the spectral reflectance model.The determination coefficients R2 of the verification models in the three key periods of jointing,heading and filling are all greater than 0.7,and the root mean square errors RMSE are less than 6%,And the relative errors RE are less than 9%.(3)An optimal monitoring model for canopy SPAD values in winter wheat was constructed.The bands sensitive to winter wheat canopy SPAD values are green light band with a central wavelength of 560 nm,red light band with a central wavelength of 668 nm,and red edge band with a central wavelength of 717 nm.Among the models established based on the univariate and stepwise regression analysis methods,the stepwise regression model has the best effect.The four vegetation indices(MSR,CARI,NDGI,TVI)were selected at the jointing stage for the best modeling effect.The R2 of the model was determined to be 0.73.The R2,RE,and RMSE of the model are 0.63,2.83%,and 1.68%,respectively.Three vegetation indices(GNDVI,GOSAVI,and CARI)were selected at the heading stage for best modeling results.The R2 of the model was determined to be 0.81,and the R2 and RE of the model were verified.And RMSE are 0.63,2.83%,and 1.68%respectively.Two vegetation indices(MSR,NDGI)are selected for the best modeling effect during the grouting period.The R2 of the calibration model is 0.67,and the R2,RE,and RMSE of the verification model are 0.65 and 2.83%,1.88%.