| Cultivating high-yielding,stable,and high-quality wheat varieties under drought conditions is currently an urgent and important challenge.The traditional method of drought-resistant wheat breeding is based on the breeder’s investigation of wheat agronomic traits,which has subjective limitations,low accuracy,and high human and material costs.Given the limitations of traditional breeding methods,there is an urgent need to develop a new method that is efficient and accurate in identifying drought tolerance in wheat plants and to explore drought-tolerant genes.Therefore,this study used computer vision technology to replace manual measurements for selecting drought-resistant wheat strains and used image feature parameters for whole-genome association analysis to explore candidate genes for drought resistance in wheat,providing valuable germplasm resources for genetic improvement of wheat drought resistance.The main contents of this study are as follows:a natural population composed of 200 wheat germplasms was selected,and wheat was planted using a pot planting method.The experiment was designed with a drought group and a control group,with three biological replicates in each group.The soil moisture content was maintained at 15-20%in the drought group and 25-30%in the control group until full maturity.During the experiment,20 agronomic traits such as plant height and tiller number were manually recorded at maturity,and the average value of each trait was calculated as the phenotype value of the manual agronomic trait.The high-throughput phenotype acquisition system was used to collect side-view images of the entire growth period of potted wheat,and image feature parameters of the potted wheat were calculated.The SE-Res Net50 network was used to classify potted wheat for drought resistance.Whole-genome association analysis was performed based on image feature parameters combined with resequencing data to explore drought-resistant genes in wheat.The main research results are as follows:(1)Study on artificial agronomic traits of wheat.The 20 agronomic traits and drought resistance coefficient obtained in this study conform to continuous normal distribution.The variation of phenotypic data in the control group and the drought group is as follows:number of grains per plant(KNP)(27.02%in the control group,50.84%in the drought group),number of fertile spikelets per plant(FSPP)coefficient of variation(27.29%in the control group,48.83%in the drought group),number of fertile spikelets per plant(FSP)(15.41%in the control group,31.55%in the drought group),and number of kernels per spike(KN)(25.04%in the control group,40.47%in the drought group).The above traits show high variability under different treatment groups,indicating that KNP,FSPP,FSP,and KN have significant phenotypic variations under drought conditions.The traits with high coefficient of variation of drought resistance coefficient are number of sterile spikelets per plant(SSPP,74.83%),grain yield per plant(GY,61.71%),and number of sterile spikelets per plant(SSP,50.84%).Under drought conditions,the above agronomic traits significantly affect the final yield of wheat.(2)Yield and biomass regression analysis based on image feature parameters.Wheat plants and ears were segmented using image segmentation algorithms and deep learning algorithms,and image feature parameters of the whole wheat plant and ear were calculated and analyzed using RGB images of potted wheat during the whole growth period.The results show that when only the total projected area(TPA)is used to predict yield and biomass,the predictive correlation coefficients R~2reach 0.787(biomass)and0.670(yield),respectively.When all image feature parameters are used to predict yield and biomass,the predictive correlation coefficients R~2reach 0.788(biomass)and 0.716(yield),respectively.Based on the predicted yield and biomass,drought resistance coefficients are calculated,and the predictive results of drought resistance coefficients for biomass and yield reach R~2values of 0.7212 and 0.6788,respectively.The results show that the image feature parameters used in this study can predict the biomass and yield and drought resistance coefficients of wheat in the early growth stage,which can provide an efficient new method for wheat drought resistance screening and identification.(3)This study focuses on the classification of wheat drought resistance levels based on deep learning classification networks using RGB images of wheat at the seedling-tillering,jointing,and heading stages.A dataset of 7,011 RGB images of wheat at the side view was used to train the model,and drought resistance levels were divided into five grades from 1 to 5.The SE-Res Net50 model with a fusion attention mechanism was found to have the best performance in predicting wheat drought resistance levels,achieving an accuracy of 98.3%on the test set.By combining the model’s predicted results with manually recorded agronomic traits,the mean drought resistance coefficient(MV＿DR)was proposed,which can improve the identification rate of wheat varieties with drought resistance.(4)Genome-wide association analysis based on image feature parameters.In this study,image feature parameters of 200 wheat natural populations were extracted,and genome-wide association analysis(GWAS)of wheat was conducted on the obtained image feature parameters.A total of 1320 SNPs significantly associated with drought resistance were identified,among which 51 SNPs associated with drought resistance were found on the image feature parameters of TPA,YPA,YTR,MU3＿TEX,and M＿TEX,which had been reported before.The analysis results show that the image feature parameters obtained in this study can not only locate SNPs related to wheat reported in previous studies but also efficiently mine new SNPs,providing valuable genetic resources for wheat drought resistance genetic improvement. |
