| Wheat stay green(SG)is an important phenotype related to wheat yield and environmental adaptability.The phenotyping for SG is a necessary and key task for revealing the SG genetic basis in wheat breeding.The spectral imaging platforms are considered effective phenotyping tools and have been widely used to accelerate crop phenotyping.It is meaningful to explore sensing techniques and analysis methods for SG phenotyping of wheat germplasm with spectral imaging means to provide support for tool selection and analysis framework.It is helpful to provide construction reference for high-throughput phenotype in wheat SG gene mapping.The work attempted to detect and analysis the SG trait in wheat germplasm and explore the explanatory power of wheat SG on yield in the field using UAV platform equipped with RGB,multispectral and thermal infrared imaging sensors to provide useful reference for high throughput SG phenotyping.The main results were as follows:(1)Analysis of classification model of wheat SG stages based on RGB and multispectral image indices.Different indices of wheat at flowering stage,watery ripe stage,milk ripe stage and wax ripe stage were extracted,and the detection models of SG stage were established respectively by using RGB color indices(CIs),multispectral CIs,multispectral spectral indices(SIs),and multispectral CIs fusion SIs to analyze their detection effects.The different SG stages could be detected by CIs extracted from the RGB and multispectral images and the detection performance of CIs derived from RGB images is better than that of CIs derived from multispectral images.The performances of SIs are better than that of CIs in RGB/multispectral images.The reflectence information of red edge or near-infrared band is more critical and effective in wheat SG detection.(2)The dynamic of RGB/multispectral indices in wheat canopy during SG change.The change trends of canopy image indices from heading completion to maturity of wheat were analyzed using high time series data.Norm R,Norm G,NGRDI,GLI,GR,VARI and Ex R in RGB and multispectral images showed monotonous changes with wheat senescence and they can be used to track the changes of wheat canopy color,but vulnerable to changes in external environment.The NDRE,NDVI,GNDVI,BNDVI,NDREI,CIRE,ARI1,ARI2 and OSAVI of the multispectral images can stably track the late development of wheat,and are less affected by external environmental factors.The redundancy between CIs is higher than that between SIs.Spectral indices could reveal the dynamics of canopy activity and pigment related traits during wheat grain filling or senescence,and CIRE,ARI1 and ARI2 could reveal canopy early senescence.(3)Quantification and identification of wheat SG based on RGB and multispectral temporal indices.The relative SG ratio(SGR)of wheat was calculated by using the temporal data of selected monotonic RGB and multispectral indces to quantify and compare wheat SG.The temporal data of SG＿Norm R,SG＿NGRDI,SG＿GLI and SG＿GR calculated by CIs and SG＿NDRE,SG＿NDVI,SG＿GNDVI,SG＿NDREI,SG＿CIRE,SG＿ARI1,SG＿ARI2 and SG＿OSAVI calculated by SIs could quantify the SG dynamic.The SG＿Norm R in RGB imaging had no effective statistical difference among samples with different SG grades during the flowering and grain-filling stages.The SG＿NGRDI,SG＿GLI and SG＿GR in RGB/multispectral imaging had effective statistical difference among samples with different SG grades in the late grain-filling stages.The SG＿NDRE,SG＿NDVI,SG＿GNDVI,SG＿NDREI,SG＿CIRE,and SG＿OSAVI had effective statistical difference among SG grades samples in the flowering and different filling stages.The SG＿NGRDI,SG＿GLI and SG＿GR at the later grain filling stages in RGB/multispectral imaging had low accuracy and the SG＿NDRE,SG＿NDVI and SG＿GNDVI had high accuracy in their respective classification model of SG grades.It needs to consider genotypes sources or other agronomic traits when identifying wheat SG with SG＿NDRE,SG＿NDVI and SG＿GNDVI in the germplasm with diversified genotypes.(4)Screening and identification for SG in wheat germplasm with thermal infrared imaging.We analyzed the midday canopy temperature(CT)at different stages of the wheat population and proposed a strategy to screen wheat SG materials based on CT ranking in the absence of the informationin related to SG.And the CT difference of samples with different SG grades was analyzed to determine the SG identification potential of thermal infrared imaging.The midday CT of wheat acquired by TIR was consistent with the field air temperature and was easily affected by external environmental factors.The CT differences of wheat samples can be well presented by TIR when the air temperature is relatively high.The strategy for ranking wheat samples CT at noon could be used to screen SG materials in wheat quickly and conveniently at flowering and grain filling stages even in the absence of SG information.The relatively low CT could be used as a valuable indicator for wheat materials with better SG performance.When the air temperature is high,the CT of good SG grade samples is significantly lower than that of medium SG grade samples,and the CT of medium SG grade samples is also lower than that of poor SG grade samples.The TIR could provide more timely auxiliary reference for selection and identification of germplasm SG types.(5)The relationship analysis of temporal SIs and wheat yield using pearson correlation coefficients(PCCs)and machine learning regression methods.The NDRE,GNDVI,CIRE and ARI2 showed extremely positive linear correlation with the yield at flowering and early filling stage and these indices could be directly used to screening high-yield materials at the early stages.The correlation between temporal SIs from flowering to maturity and yield were higher than that of single stage indice.The temporal data of NDRE,NDVI,GNDVI,CIRE,ARI2 and OSAVI have high explanatory power to the final yield and could account for 77.1%～84.4%and 75.5%～87.2% of yield variation in calibration set and prediction set,respectively.The fused data of different temporal indices could improve the explanatory power of yield and could account for more than 82.8% of yield variation in calibration and prediction set.(6)The relationship analysis of spectral temporal SGR and CT with wheat yield using PCCs and machine learning regression methods.The SG＿NDRE,SG＿NDVI,SG＿GNDVI and SG＿OSAVI could showed extreme or strong linear positive correlation with the yield in the early grain filling stages.The temporal SGR could obtain high correlations with wheat yield in regression models and the correlations between temporal SGR and yield were higher than that of single-stage SGR.The temporal data of SG＿NDRE,SG＿NDVI,SG＿GNDVI,SG＿CIRE,SG＿ARI2 and SG＿OSAVI from flowering to maturity have good explanatory power for the final yield and could account for 75.5%～83.2% and 72.9%～80.1% of final yield variation in calibration set and prediction set,respectively.The fusion of temporal SGR data can effectively improve the explanatory power of final yield and could account for more than82.8% of yield variation in calibration and prediction set.In case of high air temperature,there could be a strong or moderate linear negative correlation between wheat CT and yield in the flowering and filling stages,and the lower CT could serve as a valuable reference for screening high-yield materials.The CT fusion data from flowering to maturity have acceptable explanatory power for the final yield,and could account for 66.1%～73.4% and 71.2%～71.9%of yield variation in calibration set and prediction set,respectively. |
