Extracting Plant Height Of Crop Based On Visible Spectral Image Of UAV

The rapid and accurate acquisition of the plant height is of great significance for the precise irrigation of crop.During April to August in 2018,at the precision irrigation test base in Zhaojun Town,Dalate Banner,Inner Mongolia and the winter wheat experimental field of Institute of Soil and Water Conservation,Northwest A&F University,Yangling Demonstration Area,Shaanxi Province,obtained high-definition digital image by using the UAV visible spectral image acquisition system and measured ground plant height of maize from the emergence period to the seeding-silking period.Three-dimensional coordinates of ground control points(GCP)evenly arranged on the ground were measured by a high-precision differential GPS(global positioning system).Firstly,the digital surface model(DSM)and the digital orthophoto map(DOM)of field maize in different growth stages were established based on high-definition digital images and GCP.the maize plant height information in the digital surface model was extracted based on the binarization classification results,and the ground measured value was used to evaluate the maize plant height extraction effect of different growth stages.The main work and conclusions of this paper are as follows:(1)Visible image data of experiment area are acquired by UAV visible image acquisition system.Pix4DMapper software is used to stitch several high-resolution visible images.Geometric correction and radiation correction are made by ground control points and gray board.Arc map software is used to clip stitching image combined with surface vector files of experiment area.The final result is the whole experiment area.high-precision orthophoto and digital surface model in the domain.(2)Nine classification models were constructed by means of image preprocessing methods such as mean filtering,Gauss filtering and median filtering,maximum likelihood method,artificial neural network and support vector machine.Maize vegetation and other objects in DOM of orthophoto Image of the experiment area were classified.The classification accuracy and Kappa coefficient tables were generated by visual interpretation to classify the results.The results show that the classification model of Gauss filter-support vector machine has the best classification results in different periods(Kappa coefficient is better than 0.84,total classification accuracy is better than 91%).(3)The classification results of maize vegetation in DOM of orthophoto Image in each period were binarized by the Gauss filter-support vector machine classification model,and the vector mask file with geographic location information was established.The mask file was applied to DSM.After extracting the elevation information of Maize with the mask statistical tool,the height of Maize in different periods was calculated by the difference calculation with the ground elevation in bare soil period.Through correlation analysis with the measured plant height on the ground,the effect of corn plant height extracted by this method in different water treatment areas in different periods was evaluated.The results showed that the extraction effect at jointing stage was the best,and the determination coefficients were all greater than 0.8;the extraction effect at heading and silking stage was the second,and the determination coefficients were all better than 0.6;the extraction effect at bell stage was the worst,with the determination coefficients ranging from 0.29 to 0.5;After evaluating the extraction effect of plant height at different growth stages of all maize in the region,we can see that the extraction effect at jointing stage is the best,the determination coefficient is 0.84;the extraction effect at male drawing and silking stage is the second,the determination coefficient is 0.66;and the extraction effect at bell mouth stage is the worst,the determination coefficient is 0.63.Therefore,the effect of using UAV visible light to extract maize plant height is different in different irrigation amount and growth period.(4)Linear regression fitting of plant height extraction value and ground observation value of maize in different water treatment areas and different growth stages was carried out.A linear regression model based on DSM was established.The determination coefficient of the model was 0.89,the determination coefficient of the verification set was 0.91,and the height of the model was fitted(p<0.01).The extraction effect of plant height is good.It is feasible to estimate the true value of plant height by using the plant height extracted from UAV visible image.(5)Using this method to obtain high-definition digital images of wheat from elongation stage to maturation stage and set up crop digital surface model and crop height model.The research shows that CHM crop height during different growth period is significantly related to the actual crop height(P<0.01,R~2=0.8200,RMSE=0.0431),the range of error is-0.10m-0.09m,the average relative error is 17.64%-19.60%.It is feasible to use the UAV to obtain high-definition digital images which helps to estimate the wheat height quickly and accurately.This study can provide a reference for rapid and accurate detection of agricultural conditions in large areas.