Study On High-Throughput Maize Phenotyping Analysis And Evaluation Based On UAV Quantitative Remote Sensing

With environmental deterioration,natural resource scarcity,and rapid population growth,mankind is facing severe global food security problems.To meet future needs,it is necessary to accelerate progress in breeding for new varieties with high yield and strong resistance.However,the traditional phenotypic screening methods have some disadvantages,such as destructive,inefficient,low-dimensional,labor-intensive and cumbersome,which seriously hinder the development of field breeding.Breeders urgently need a high-throughput technique for acquiring and evaluating phenotypic data that can efficiently and easily screen out excellent phenotypic traits from large-scale genotype populations.Because of its real-time,multi-source and dynamic characteristics,UAV remote sensing technology provides a new solution for high-throughput acquisition and evaluation of crop phenotypic data.In this paper,a UAV remote sensing high-throughput phenotyping platform was used to obtain high-throughput phenotypic information of large-scale maize(Zee mays L.)breeding materials,and the representative anti-resistance phenotypic traits,i.e.,lodging,morphological structural traits,i.e,plant height and production performance traits,i.e,aboveground fresh biomass,were studied in depth.By establishing phenotypic maps and phenotypic similarities in large-scale maize breeding materials,a comprehensive evaluation for multiple phenotypic traits was accomplished.The main research contents and conclusions are as follows:(1)A method for estimating plant height considering the spatial structure of maize canopy was proposed.From the experimental data,it was found that the maize-canopy structure at the plot scale changed with the growth,and the high-density point cloud distribution also had a certain randomness.Based on the above observations,a new method for estimating plant height was proposed.The core idea was to retain the top canopy structure as much as possible while removing the influence of the lower leaves.This method used the 3D coordinate information of point cloud to ensure that many plants participate in the calculation of plant height,so it can overcome the outliers to a certain extent.(2)By measuring the similarity of phenotypic expression dynamic change patterns(spatiotemporal profiles),three spatiotemporal phenotypic traits related to plant height were clustered by using the fuzzy C-means algorithm.This revealed the variation of plant height in spatiotemporal dimension,and verified that the spatiotemporal phenotypic traits obtained by UAV remote sensing high-throughput phenotypic platform can be used for phenotypic selection in the field.(3)The canopy structure and spectral information provided by UAV remote sensing were combined with machine learning method to estimate the aboveground fresh biomass on maize.Firstly,the recursive feature elimination algorithm was used to select 14 predictive variables with the minimum RMSE as the optimization objective,and the optimal subset of only 6 predictive variables was obtained.Then,four machine learning models(multiple linear regression,support vector machine,artificial neural network,and random forest)were evaluated and compared to create a suitable model.Considering the performance of the models in the training set and the test set at the same time,the random forest model gave the most balanced results,i.e.,smaller error and higher explained variance ratio(training set:R~2=0.94,RMSE=0.49 kg/m~2;test set:R~2=0.699,RMSE=1.2 kg/m~2).The predictions provided by the random forest model can reflect the difference in the above-ground fresh biomass among the genotype population.(4)A nomogram model was proposed to identify maize lodging.Firstly,several potential factors for identifying maize lodging were determined preliminarily by literature analysis.Then,the difference significance test,single factor logistic regression analysis and multivariate logistic regression analysis were used gradually to determine the predictive variables for modeling.Finally,two nomogram models were constructed and compared from distinguishing ability and robustness.The results show that the nomogram model can not only quantitatively predict the probability of lodging at the plot sacle,but also identify the risk factors and protective factors associated with maize lodging.Its visual calculation method was convenient for the promotion and application.(5)Based on phenotypic map and phenotypic similarity in large-scale maize breeding materials,several phenotypic traits including flowering time,above-ground fresh biomass,average growth rate of plant height,plant height,leaf color,lodging and yield were evaluated comprehensively.Firstly,combining the advantages of self-organizing feature map(neural networks)and hierarchical clustering,a "two-step"clustering method was proposed to construct phenotypic maps.Then,principal component analysis biplot was used to analyze the relationships among multiple phenotypic traits.Finally,based on different breeding targets,the preliminary screening for genotypic population and multi-phenotypic traits was accomplished.