Font Size: a A A

Wheat Growth Monitoring And Yield Prediction Based On The Unmanned Aerial Vehicle Platform

Posted on:2017-05-15Degree:MasterType:Thesis
Country:ChinaCandidate:N WangFull Text:PDF
GTID:2323330518478195Subject:Crop Cultivation and Farming System
Abstract/Summary:PDF Full Text Request
It is key to monitor crop growth efficiently and nondestructively and predict crop yield timely and accurately for modem precision agriculture.For crop growth monitoring and yield prediction,Unmanned Aerial Vehicle(UAV)platform is superior to ground remote sensing platform and satellite platform in many aspects,such as low price,efficient data acquisition and flexible flight altitude.In this study,three field experiments involving different wheat varieties,nitrogen application rates and planting densities were conducted during two wheat growing seasons.Wheat multispectral imagery were collected in the main critical wheat growth stages by Mini-MCA6 camera equipped on the eight-rotor Oktokopter.Simultaneously,Leaf area index,leaf biomass and grain yield were obtained through destructive field samplings.The objectives of this study were:?)to explore the image preprocessing methods for the multispectral imagery;?)to extract wheat canopy reflectance and select the optimal vegetation index to monitor wheat growth status and predict yield;?)to establish reliable models to monitor wheat growth status and predict yield.Our work might provide effective technical support for real-time monitoring of wheat growth and yield prediction in small and medium-sized regions.We set two controlled indoor environments(noise imagery were taken in a totally dark box;vignetting imagery were taken in an integral sphere with homogeneous light)and one outdoor environment(spectral data after radiation correction were collected in sunny and cloudless weather)to construct the image preprocessing workflows and compare different processing methods.Six major sensor correction components were determined:noise reduction,vignetting correction,lens distortion correction,band registration,radiation correction and image classification.Noise reduction,vignetting correction and lens distortion correction was done through the correction coefficient from the imagery.Band registration was based on ground control points and Tetracam PixelWrench2(PW2)software.Radiation correction was done through the empirical linear correction method and the light intensity sensor correction method.Maximum likelihood classification method and support vector machine were adopted to classify the multispectral imagery of the five critical stages.Calibration error and accuracy were used to evaluate the performance of different methods and the best method was selected to add into the preprocessing workflow.The results showed that band registration based on ground control points was superior to the one through the use of PW2 software.For radiation correction,the calibration error of empirical linear correction method was significantly lower than that adopting the light intensity sensor correction method.However,the two methods obtained similar performance on wheat growth indicators monitoring,with a relative error below 20%.The background was removed with the maximum likelihood method combining spectrum with texture feature,which performed best with the highest overall classification accuracy and kappa coefficient.After the image preprocessing workflows was determined,wheat canopy spectra were extracted from the imagery.Eleven commonly used near-infrared vegetation indices were calculated to established LAI,leaf biomass monitoring models and compared the model results with data collected from the ground remote sensing platform.The correlations between the vegetation indices in the different growth stages and yield were analyzed.We established multiple linear regression models between the vegetation indices from the different growth periods and yield.The highest coefficients of determination were obtained between LAI,leaf biomass and the modified triangular vegetation index(MTVI2)based on the UAV platform.The determination coefficients(R2)and relative root mean square error(RRMSE)of the monitoring model and validation model were 0.79,0.75,0.80 and 0.80,24%and 20%,respectively.Compared with the ground platform results,the monitoring effect was not affected by the flight height.The best results were obtained using the ratio vegetation index(RVI)from the heading,anthesis and filling stage to establish a multiple linear regression model for yield prediction.The R2 of the monitoring model and validation model were 0.65,0.68 and RRMSE was 15%.These results might provide effective technical support for nondestructive monitoring of wheat growth status and grain yield prediction in small and medium-sized regions.
Keywords/Search Tags:UAV, Multispectral imagery, Image processing methods, LAI, Leaf biomass, Yield
PDF Full Text Request
Related items