Research And Application System Development Of Maize Canopy Image LAI Estimation Based On Computer Vision

Maize leaf area index(LAI)is an important parameter to characterize the growth status of maize crops at seedling stage,and it plays an important role in evaluating maize population quality,accurate nitrogen fertilizer management and yield prediction.Therefore,non-destructive,efficient and rapid acquisition of LAI in field environment is of great significance for maize seedling growth monitoring and fine farmland management.In this paper,smart phone and LAI-2200 C plant canopy analyzer were used to collect canopy image data and LAI measured values of maize breeding materials at jointing stage,small trumpet stage and big trumpet stage(hereinafter collectively referred to as maize seedling stage).Three leaf area index estimation models were proposed based on computer vision technology.A variety of image features were extracted from maize canopy images,which were taken as independent variables of the three algorithm models,and the measured values of LAI were used as dependent variables to establish the relationship between maize canopy images and leaf area index,and the accurate estimation of LAI from jointing stage to trumpet stage of maize breeding materials in field environment was realized.Based on image processing technology and convolution neural network,the main research contents and results of this paper are as follows:1.construct linear regression LAI estimation model.Related studies show that there is a good linear relationship between maize seedling image canopy coverage(CC)and LAI parameters,so the linear regression model(LR)with CC as independent variable is used as one of the estimation methods.Canopy coverage is obtained by calculating the ratio of vegetation pixels to the total pixels in maize canopy images at seedling stage.The research data are divided into test set and training set,and a LR estimation model is established,in which the input is the image canopy coverage and the output is the current maize seedling population leaf area index.2.construct the LAI estimation model of support vector regression.A large number of global features of maize canopy images were extracted from maize seedling canopy images,and the research data were divided into test set and training set according to the same proportion as LR estimation model.The relationship between image color features,texture features and maize leaf area index in the same period was obtained by using support vector regression(SVR)algorithm.The global image features are extracted from the canopy image of maize population at seedling stage.The extracted features include nine color components in three color spaces: RGB,HSV and L*a*b*,including color features such as first moment(Avg),second moment(Std)and texture features such as contrast(Contrast),homogeneity(Homogeneity)and so on.After feature extraction,the Pearson correlation analysis is used to select the features with high correlation with the estimated parameters to construct the SVR estimation model.3.construct the LAI estimation model of convolution neural network.Taking the visible light image of maize canopy at seedling stage as input,a leaf area index convolution neural network(CNN)estimation model suitable for maize from jointing stage to trumpet stage was constructed.The relationship between maize canopy image and leaf area index in the same period was established by means of automatic learning characteristics,and the accurate estimation of corn leaf area index in field environment was realized.In the study,it was found that the accuracy of CNN model decreased with the growth of maize crops,and the plant type of maize crops was higher.The "leaf overlap" of canopy plants resulted in a certain deviation between the real leaf area index and the effective leaf area index.In order to correct the deviation,the maize growth period information and the maize population canopy image are used as the input of the CNN model,which effectively improves the accuracy of the CNN estimation model.4.to develop an application system for estimating leaf area index of maize breeding materials at seedling stage.Select the optimal maize seedling LAI estimation model,use Java and Python programming language,based on Spring MVC framework technology,adopt MVC framework design pattern,take My SQL as database,Tomcat as Web container,development tools select Eclipse,to use self-coding Python script to load and calculate CNN estimation model,and construct the leaf area index estimation application system of maize breeding materials at seedling stage.The design of the application system is based on the Bachet S mode,including user management,image management,LAI estimation result recording,expert opinion management and other modules.It provides the functions of user image upload,user image management,LAI estimation results and record query based on image processing technology.In this paper,coefficient of determination(R2)and normalized root mean square error(n RMSE)are used to quantitatively evaluate the accuracy of the research method,in which the R2 of the LR estimation model is 0.7811,the n RMSE is 23.00%,the R2 of the LR estimation model is 0.5721,the n RMSE is 38.66%,the R2 of the LR estimation model is 0.894,and the n RMSE is 19.27%.The accuracy of CNN estimation model is higher than that of the other two estimation models.As the optimal model,CNN estimation model is suitable for LAI estimation from jointing stage to trumpet stage of maize.The application system of leaf area index estimation of maize breeding materials at seedling stage based on Java Web technology can provide users with field phenotypic information monitoring of maize breeding materials conveniently and quickly,which is of positive significance for guiding maize production,and provides a new solution for LAI estimation of maize breeding materials in the field.