Research On Rice Yield Estimation Based On Digital Imaging Of UAV

Rice is the main food crop in China.Scientifically and accurately predicting the yield of rice will not only provide timely and accurate information on agricultural information,but also be of great significance to the formulation of agricultural policies.At present,the main method of rice yield estimation is satellite remote sensing yield estimation,but its resolution is low,its error is large,its mechanism is not sufficient and can not be further popularized and applied.Therefore,this study use UAV platform equipped with high-definition digital camera to shoot the image of rice canopy,then extract the rice panicle by processing the digital image,and then the rice yield estimation formula is used to estimate the yield.The main research contents and results of this paper are as follows:(1)In this study,the canopy image of rice can be taken from the heading stage to the mature stage by using the UAV platform equipped with high-definition digital camera.After the image is denoised and preprocessed,the image is converted from the RGB color space to the L*a*b* color space for the color characteristics of the color rice image.K-means clustering algorithm is applied to cluster analysis and image segmentation of rice canopy images to extract rice panicles and obtain rice panicle numbers,and then estimate the yield.The images taken on different days are processed separately.Among them,clustering and segmentation are performed on the images taken at the full heading stage of rice.The effect of extracting rice panicles is better,the precision of estimating yield is higher,and the estimating root mean square error and mean absolute percentage error of yield are 9.08 and 22.8%,respectively.(2)In this study,the rice canopy images captured by unmanned aerial vehicle are classified and identified.Ten rice cultivars are initially selected to classify the panicles,rice leaves and background.The selected features included R,G,B values and projections at the pixel level of the image.H,S,V values to HSV space,and four indices: Normalized Green Red Difference Index(NGRDI),Red Green Ratio Index(RGRI),Green Leaf Index(GLI),and Excess Green(EXG)values.Then,based on the best subset selection algorithm,ten classification features are optimized.The selected classification model features are G,B,H,S,V,RGRI,and GLI.Threshold segmentation is used to segment the image and extract rice panicle for yield estimation.Processing the images taken on different dates,it is concluded that the image segmentation results obtained during full heading stage of rice are better,the accuracy of extracting rice panicles is higher,and the root mean square error and mean absolute percentage error of yield estimation are the smallest,being 6.85 and 18.2%,respectively.The results of the K-means clustering method and the best subset selection method are compared,and it is found that the best subset selection method have better results in extracting rice panicles.(3)This study select and classify the rice images captured by three-meter,six-meter and nine-meter height unmanned aerial vehicle based on the best subset selection algorithm.The initial selection features are R,G,B,H,S,V,NGRDI,RGRI,GLI and EXG,the results show that the image classification recognition error of three shooting heights decreases with the increase of the model feature number,and when the feature number is greater than seven,the error remains basically unchanged,so the three heights is seven features are selected as the optimal classification features.Then,the threshold segmentation method is used to segment the three heights of rice images,and the rice panicles are extracted for estimate yield.The results show that the accuracy of rice panicle identification at three meters height is higher than that of six meters and nine meters,and the estimate mean absolute percentage error of rice panicles reached 10.00%;the estimation accuracy of rice yield at three meters height is higher than six meters and nine meters,and the estimate mean absolute percentage error of yield reached 15.66%.