Research On The Estimation Method Of Key Parameters Of Wheat Field Phenotypes

In the process of wheat breeding,the field phenotype is not only the guiding data for germplasm selection in the early stage of breeding,but also the evaluation data for the later stage of popularizing planting,However,there are few researches on the cooperative measurement of multiple wheat varieties.There is a lack of effective high-throughput method for obtaining multi-strain crops.In this study,the research area is located in field 2,caoxinzhuang experimental farm,Northwest Agricultural and Forestry University,as the test objects are wheat breeding material.The multispectral and visible images of wheat from heading stage to maturity in 2018 and jointing stage to maturity in 2019 were obtained by using UAV multispectral camera and RGB camera.Three-dimensional point cloud model and spectral vegetation index image of the field were established.Research on feature extraction method was carried out.These features were correlated with leaf area index,biomass and chlorophyll content.According to the key features,their corresponding estimation models were established,which is helpful to explore the application of UAV Remote Sensing Technology in crop high-throughput phenotyping.The conclusions are as follows:?1?The leaf area index of wheat is related to the volume V of Three-dimensional point cloud,and the values of Chlorophyll Index-Green?CIG?,Green Normalized Difference Vegetation Index?GNDVI?,Green,Red and Red Edge of spectral vegetation index images.For the establishment of wheat leaf area index estimation model,using the volume V of three-dimensional point cloud,the combination of the values of spectral vegetation index images CIG,GNDVI,Green,Red and Red Edge,the prediction ability of the estimation model established in the heading stage of wheat is better than that in the mature stage.Compared with partial least square regression,support vector machine regression and Gaussian process regression,the prediction ability of the estimation model based on Gaussian process regression is better.Based on the regression of Gauss process,the coefficient of determination between the estimated value of the test set and the measured value is 0.879,and the root mean square error is 0.325.?2?The biomass is related to the characteristics of spectral vegetation index Chlorophyll Index-Green?CIG?,Chlorophyll Index-Red Edge?CIRE?,Normalized Difference Red Edge Index?NDRE?,Green Normalized Difference Vegetation Index?GNDVI?,Normalized Difference Vegetation Index?NDVI?,Transformed Chlorophyll Absorption and Reflectance Index?TCARI?and Triangular Vegetation Index?TVI?.For the establishment of wheat biomass estimation model,the prediction ability of using the combination of characteristic values of spectral vegetation index images CIG,CIRE,NDRE,GNDVI,NDVI,TCARI and TVI in jointing and flowering stage of wheat is better than that in heading stage and maturity stage.Compared with partial least square regression,support vector machine regression and Gaussian process regression,the prediction ability of the estimation model based on Gaussian process regression is better.Based on the Gauss process regression,the decision coefficient between the estimated value and the measured value is 0.832,and the root mean square error is 101g/m².?3?The content of chlorophyll in flag leaf of wheat is related to the characteristic values of CIG,CIRE,NDRE,GNDVI,NDVI,TCARI and TVI.For the establishment of wheat chlorophyll content estimation model,the prediction ability of using spectral vegetation index image CIG,CIRE,NDRE,GNDVI,NDVI,TCARI and TVI combination in wheat heading and flowering stage is better than that in jointing stage and maturity stage.Compared with partial least square regression,support vector machine regression and Gaussian process regression,the prediction ability of the estimation model based on Gaussian process regression is better.Based on Gaussian process regression,the decision coefficient between the estimated value and the measured value is 0.943,and the root mean square error is 4.58mg·g^-1FW.