Establishment Of Intelligence Phenotypic Platform For Wheat Roots Under Stress And Its Genetic Application

The increase in wheat production is of great strategic importance to global food security.However,the change of climate and the decrease of mineral resources lead to frequent drought,phosphorus deficiency,high temperature and other abiotic stress environments.Rapid identification of wheat cultivars with stress tolerance and analysis of genetic basis of them are important for wheat stress tolerance breeding.With the development of remote sensing,robotics,computer vision and artificial intelligence,plant phenotypic omics provides a new research idea for wheat stress tolerance breeding,which combines phenotypic information with machine learning technology to complete the prediction of wheat stress tolerance indexes and the identification of wheat varieties with stress tolerance.In this study,we took wheat genetic population phenotypes under phosphorus deficiency and drought stress as examples to collect multi-source data on root traits.Combined with machine learning algorithm,the phenotype analysis of wheat under stress was carried out to realize the classification of stress-tolerant varieties and the prediction of stress tolerance ability.Based on the deep learning technology,the wheat root image was analyzed to achieve the efficient recognition of wheat stress-tolerant varieties.To establish the wisdom phenotype platform of wheat root system under stress stress,and provide the function of wheat stress tolerance analysis.To improve the automation and intelligence level of wheat stress phenotype analysis,and provide key technical support for wheat stress phenotype omics research.The main results obtained are as follows:1.The data sets and photo sets of wheat phenotype of phosphorus deficiency stress and drought stress tolerance were established,which enriched the data basis of wheat phenotype omics for stress tolerance breeding.Using the SHW-L1/CM32 RILs and wheat landraces,17 stress sensitivity index(SSI)related traits and 6083 original root line images under phosphorus deficiency and drought stress conditions were obtained by hydroponic method.The establishment of large data sets lays a foundation for wheat stress tolerance prediction and recognition based on machine learning and deep learning methods.In addition,using the comprehensive stress tolerance sensitive index D value(D.SSI)as the evaluation index,a total of 21 phosphorus deficiency resistant materials and 16 drought resistant materials were selected from the two populations,which enriched the germplasm resources of wheat stress tolerance.2.The application value of random forest algorithm in the field of wheat stress prediction was explored,and the prediction model of wheat stress tolerance level and ability was established,which laid the technical foundation for the big data phenotype analysis of wheat stress.Based on the phenotypic data set constructed in the early stage and random forest algorithm,the phenotypic analysis of wheat under stress conditions was completed.The results showed that the random forest algorithm could complete the evaluation of the importance of variables quickly and effectively.And the D-value prediction regression model and variety identification model of wheat under phosphorus deficiency stress and drought stress were constructed.Among them,there are two models related to phosphorus deficiency stress of wheat,which are the regression model "D.PSSI model B" with 91.56%fit degree and the classification model "P deficiency stress level model B" with 6.88%out-of-bag error.There are two drought stress-related models for wheat,which are the regression model "D.SSI model B" with 86.9% fit and the classification model "drought stress tolerance grade model B" with 4.64% out-of-bag error.Combined with the random forest algorithm,a large number of characteristic variables or a few important characteristic variables can be used to abtain high accuracy classifiers.The data showed that the regression model constructed in this study could successfully predict the stress tolerance value of wheat,and the regression model could successfully predict the stress tolerance level of wheat.3.The application value of deep learning technology in the field of wheat stress prediction was explored,and the image recognition model of wheat varieties with phosphorus deficiency stress was established,which provided a new method for the identification of wheat varieties with phosphorus deficiency stress,and accelerated the identification process of varieties with phosphorus deficiency stress.Deep learning technology was used to construct an image set of wheat under phosphorus deficiency stress,which enriched digital agricultural data resources.The image recognition model of wheat cultivars with phosphorus deficiency stress was constructed based on simple convolutional neural network,complex convolutional neural network and VGG16 migration pre-training network by using the previously constructed image set and deep learning technology.Among them,the VGG16 pre-training network model had the highest applicability and could better complete the task of identifying wheat sensitive to phosphorus deficiency stress,with the accuracy of 88.18% in the test set and 72.60% in the validation set.4.Develop the intelligent phenotypic platform of root system with drought stress and phosphorus deficiency stress,and lay the software foundation for the realization of precision agriculture.In this study,the Shiny package of R language and the related R packages were used to develop a Web program for the identification of cultivars with drought stress and drought stress tolerance.The regression models "D.PSSI model B","D.DSSI model B",classification models "P deficiency stress level model B","drought stress level model B" and phosphorus deficiency stress tolerance variety recognition model based on VGG16pre-training network were embedded to develop the intelligent phenotype platform.The establishment of the platform realized the rapid identification and ability prediction of stress-tolerant varieties,and the results could be applied to the identification of QTL.The establishment of the platform laied the software foundation for the realization of precision agriculture.